| { |
| "paper_id": "2021", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T10:53:48.890665Z" |
| }, |
| "title": "Cryptocurrency Day Trading and Framing Prediction in Microblog Discourse", |
| "authors": [ |
| { |
| "first": "Anna", |
| "middle": [], |
| "last": "Paula", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Pawlicka", |
| "middle": [], |
| "last": "Maule", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "apmaule@gmail.com" |
| }, |
| { |
| "first": "Marie", |
| "middle": [], |
| "last": "Kristen", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "kristenj@msu.edu" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Johnson", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Eric", |
| "middle": [], |
| "last": "Baumer", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Elisha", |
| "middle": [], |
| "last": "Elovic", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Ying", |
| "middle": [], |
| "last": "Qin", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Francesca", |
| "middle": [], |
| "last": "Pol", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Johan", |
| "middle": [], |
| "last": "Bollen", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Huina", |
| "middle": [], |
| "last": "Mao", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Alberto", |
| "middle": [], |
| "last": "Pepe", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Amber", |
| "middle": [ |
| "E" |
| ], |
| "last": "Boydstun", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Dallas", |
| "middle": [], |
| "last": "Card", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Justin", |
| "middle": [ |
| "H" |
| ], |
| "last": "Gross", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Lauren", |
| "middle": [ |
| "M" |
| ], |
| "last": "Burch", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Evan", |
| "middle": [ |
| "L" |
| ], |
| "last": "Frederick", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Ann", |
| "middle": [], |
| "last": "Pego", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Ali", |
| "middle": [], |
| "last": "Derakhshan", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Hamid", |
| "middle": [ |
| "2019" |
| ], |
| "last": "Beigy", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Sentiment", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Anjalie", |
| "middle": [], |
| "last": "Field", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Doron", |
| "middle": [], |
| "last": "Kliger", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Shuly", |
| "middle": [], |
| "last": "Wintner", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Amir", |
| "middle": [], |
| "last": "Hadifar", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Lucas", |
| "middle": [], |
| "last": "Sterckx", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Thomas", |
| "middle": [], |
| "last": "Demeester", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "With 56 million people actively trading and investing in cryptocurrency online and globally in 2020, there is an increasing need for automatic social media analysis tools to help understand trading discourse and behavior. In this work, we present a dual natural language modeling pipeline which leverages language and social network behaviors for the prediction of cryptocurrency day trading actions and their associated framing patterns. This pipeline first predicts if tweets can be used to guide day trading behavior, specifically if a cryptocurrency investor should buy, sell, or hold their cryptocurrencies in order to make a profit. Next, tweets are input to an unsupervised deep clustering approach to automatically detect trading framing patterns. Our contributions include the modeling pipeline for this novel task, a new Cryptocurrency Tweets Dataset compiled from influential accounts, and a Historical Price Dataset. Our experiments show that our approach achieves an 88.78% accuracy for day trading behavior prediction and reveals framing fluctuations prior to and during the COVID-19 pandemic that could be used to guide investment actions. 1 Introduction Beginning with the 2008 introduction of Bitcoin (BTC) (Nakamoto, 2008), a cryptocurrency for a Peer-to-Peer cash system, the use of cryptocurrencies and their corresponding blockchains have increasingly gained in popularity. In 2019, the number of Americans owning cryptocurrency doubled from 7% in 2018 to 14%, representing about 35 million people trading and investing with cryptocurrency (Partz, 2019).", |
| "pdf_parse": { |
| "paper_id": "2021", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "With 56 million people actively trading and investing in cryptocurrency online and globally in 2020, there is an increasing need for automatic social media analysis tools to help understand trading discourse and behavior. In this work, we present a dual natural language modeling pipeline which leverages language and social network behaviors for the prediction of cryptocurrency day trading actions and their associated framing patterns. This pipeline first predicts if tweets can be used to guide day trading behavior, specifically if a cryptocurrency investor should buy, sell, or hold their cryptocurrencies in order to make a profit. Next, tweets are input to an unsupervised deep clustering approach to automatically detect trading framing patterns. Our contributions include the modeling pipeline for this novel task, a new Cryptocurrency Tweets Dataset compiled from influential accounts, and a Historical Price Dataset. Our experiments show that our approach achieves an 88.78% accuracy for day trading behavior prediction and reveals framing fluctuations prior to and during the COVID-19 pandemic that could be used to guide investment actions. 1 Introduction Beginning with the 2008 introduction of Bitcoin (BTC) (Nakamoto, 2008), a cryptocurrency for a Peer-to-Peer cash system, the use of cryptocurrencies and their corresponding blockchains have increasingly gained in popularity. In 2019, the number of Americans owning cryptocurrency doubled from 7% in 2018 to 14%, representing about 35 million people trading and investing with cryptocurrency (Partz, 2019).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "the pandemic's effect on Wall Street (i.e., the New course and its effects on public opinion has been 116 widely studied in NLP (Ritter et al., 2010; Walker 117 et al., 2012; Abu-Jbara et al., 2013; Hasan and 118 3 https://github.com/MSU-NLP-CSS/crypto-framing Ng, 2014; West et al., 2014; Sridhar et al., 2015) and the social sciences (Bollen et al., 2011; Harlow and Johnson, 2011; Meraz and Papacharissi, 2013; Burch et al., 2015; Jang and Hart, 2015) .", |
| "cite_spans": [ |
| { |
| "start": 128, |
| "end": 149, |
| "text": "(Ritter et al., 2010;", |
| "ref_id": "BIBREF21" |
| }, |
| { |
| "start": 150, |
| "end": 174, |
| "text": "Walker 117 et al., 2012;", |
| "ref_id": null |
| }, |
| { |
| "start": 175, |
| "end": 198, |
| "text": "Abu-Jbara et al., 2013;", |
| "ref_id": null |
| }, |
| { |
| "start": 199, |
| "end": 199, |
| "text": "", |
| "ref_id": null |
| }, |
| { |
| "start": 262, |
| "end": 271, |
| "text": "Ng, 2014;", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 272, |
| "end": 290, |
| "text": "West et al., 2014;", |
| "ref_id": null |
| }, |
| { |
| "start": 291, |
| "end": 312, |
| "text": "Sridhar et al., 2015)", |
| "ref_id": "BIBREF25" |
| }, |
| { |
| "start": 337, |
| "end": 358, |
| "text": "(Bollen et al., 2011;", |
| "ref_id": null |
| }, |
| { |
| "start": 359, |
| "end": 384, |
| "text": "Harlow and Johnson, 2011;", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 385, |
| "end": 414, |
| "text": "Meraz and Papacharissi, 2013;", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 415, |
| "end": 434, |
| "text": "Burch et al., 2015;", |
| "ref_id": null |
| }, |
| { |
| "start": 435, |
| "end": 455, |
| "text": "Jang and Hart, 2015)", |
| "ref_id": "BIBREF4" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "There are many works on Twitter sentiment analysis, but closest to our work are those concerning the use of Twitter sentiment for stock market predictions (Kouloumpis et al., 2011; Rao and Srivastava, 2012; Si et al., 2013; Derakhshan and Beigy, 2019) .", |
| "cite_spans": [ |
| { |
| "start": 155, |
| "end": 180, |
| "text": "(Kouloumpis et al., 2011;", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 181, |
| "end": 206, |
| "text": "Rao and Srivastava, 2012;", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 207, |
| "end": 223, |
| "text": "Si et al., 2013;", |
| "ref_id": "BIBREF24" |
| }, |
| { |
| "start": 224, |
| "end": 251, |
| "text": "Derakhshan and Beigy, 2019)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "There are relatively few works concerning cryptocurrency analysis and prediction. Of these, a majority use social media sentiment (Jain et al., 2018; , volume of tweets (Vidal, 2020) , or both (Abraham et al., 2018) as the main feature for prediction. Furthermore, the prediction tasks are typically to predict prices or whether those prices will rise or fall. However, sentiment is known to be difficult to predict on Twitter. Furthermore, the volume of tweets can be falsely inflated by bots reporting currency prices, but not contributing to the discourse. Therefore, instead of sentiment or tweet volume, we aim to use the language directly extracted from tweets, their context, and features representing the social network behavior for a buy, sell, or hold investment action prediction.", |
| "cite_spans": [ |
| { |
| "start": 130, |
| "end": 149, |
| "text": "(Jain et al., 2018;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 169, |
| "end": 182, |
| "text": "(Vidal, 2020)", |
| "ref_id": "BIBREF28" |
| }, |
| { |
| "start": 193, |
| "end": 215, |
| "text": "(Abraham et al., 2018)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Research has also shown that given an adequate amount of historical data, such as stock values and indices, it is possible to forecast future currency exchanges (Walczak, 2001) . Different from this work, we focus on predicting cryptocurrency investment actions, instead of fiat currency prices, by extracting patterns from historical tweets rather than stock values and indices.", |
| "cite_spans": [ |
| { |
| "start": 161, |
| "end": 176, |
| "text": "(Walczak, 2001)", |
| "ref_id": "BIBREF29" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Previous works have shown the effectiveness of using frames to predict various social sciences phenomena, such as political framing of Twitter discourse, congressional speeches, and news coverage of current events (Boydstun et al., 2014; Baumer et al., 2015; Card et al., 2015; Tsur et al., 2015; Jang and Hart, 2015; Fulgoni et al., 2016; Johnson et al., 2017; Field et al., 2018) . Framing has also been used to understand the role of Twitter discussions in influencing public opinion of events such as riots and protests (Harlow and Johnson, 2011; Meraz and Papacharissi, 2013; Burch et al., 2015) .", |
| "cite_spans": [ |
| { |
| "start": 214, |
| "end": 237, |
| "text": "(Boydstun et al., 2014;", |
| "ref_id": null |
| }, |
| { |
| "start": 238, |
| "end": 258, |
| "text": "Baumer et al., 2015;", |
| "ref_id": null |
| }, |
| { |
| "start": 259, |
| "end": 277, |
| "text": "Card et al., 2015;", |
| "ref_id": null |
| }, |
| { |
| "start": 278, |
| "end": 296, |
| "text": "Tsur et al., 2015;", |
| "ref_id": "BIBREF26" |
| }, |
| { |
| "start": 297, |
| "end": 317, |
| "text": "Jang and Hart, 2015;", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 318, |
| "end": 339, |
| "text": "Fulgoni et al., 2016;", |
| "ref_id": null |
| }, |
| { |
| "start": 340, |
| "end": 361, |
| "text": "Johnson et al., 2017;", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 362, |
| "end": 381, |
| "text": "Field et al., 2018)", |
| "ref_id": null |
| }, |
| { |
| "start": 524, |
| "end": 550, |
| "text": "(Harlow and Johnson, 2011;", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 551, |
| "end": 580, |
| "text": "Meraz and Papacharissi, 2013;", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 581, |
| "end": 600, |
| "text": "Burch et al., 2015)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Despite this coverage, to the best of our knowledge we are the first to study the role of framing in economics, specifically concerning stocks or cryptocurrency day trading, or associated correlations with the current pandemic. This work presents a first step in understanding both cryptocurrency day trading and how framing can reveal insights about cryptocurrency trading.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "This section describes the collection and preprocessing steps of the tweets and historical Bitcoin (BTC) transaction prices. Section 3.4 describes how tweets were annotated for use in the day trading behavior prediction model. The non-annotated version of these tweets were used in the framing clustering models.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Data Annotation", |
| "sec_num": "3" |
| }, |
| { |
| "text": "For this work we collected tweets related to cryptocurrency, including BTC and other popular coin types such as Etheureum (ETH) and XRP, because prices of different cryptocurrencies are highly correlated (Magas, 2020) . Rather than collect based on hashtags or keywords alone, we narrowed our search to specific time frames and user accounts.", |
| "cite_spans": [ |
| { |
| "start": 204, |
| "end": 217, |
| "text": "(Magas, 2020)", |
| "ref_id": "BIBREF14" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Tweets were scraped from January 2017, when Bitcoin surpassed $1,000 per coin, until February 2020. This time range covers times of frequent changes in cryptocurrency trading and adheres to the finding that an optimal dataset for financial time series prediction consists of information from the past two years (Walczak, 2001 ). These tweets form our Pre-COVID Dataset.", |
| "cite_spans": [ |
| { |
| "start": 311, |
| "end": 325, |
| "text": "(Walczak, 2001", |
| "ref_id": "BIBREF29" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Within these time frames, three types of user accounts were identified for tweet collection (details in A.1) to maximize the presence of discourse for analysis and minimize tweet noise. These include influential cryptocurrency Twitter accounts, or influencers, which are well known as sources for investment information and thus should provide features for message propagation. This category also includes users who frequently tweet about cryptocurrency and have at least ten thousand followers. Similarly, media accounts from traditional or online news sources, such as @CNNBusiness and @BitcoinMagazine, are used. Lastly, we include company accounts, e.g., @IBMBlockchain and @BitPay. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "If the momentum on a given day increases or decreases by 5% on the following day, then we label tweets of that given day as buy or sell, respectively.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "If there is less than a 5% change, these tweets are neutral in terms of buying or selling, and are therefore labeled as hold, to represent that an investor should take no action with their cryptocurrency.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "This 5% cutoff was chosen because BTC volatility during 2017 and 2018 was around 8% (Reiff, 2020) , while between 2019 and 2020 it was 4.66% (Tuwiner, 2020) . For comparison, the average day trading volatility of stocks is 3.3%, which is considered to be high (Kyr\u00f6l\u00e4inen, 2008) . This annotation was automated with a Python script that cross referenced the date of the tweet with the BTC Historical Price Dataset and is dependent only on price data.", |
| "cite_spans": [ |
| { |
| "start": 84, |
| "end": 97, |
| "text": "(Reiff, 2020)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 141, |
| "end": 156, |
| "text": "(Tuwiner, 2020)", |
| "ref_id": "BIBREF27" |
| }, |
| { |
| "start": 260, |
| "end": 278, |
| "text": "(Kyr\u00f6l\u00e4inen, 2008)", |
| "ref_id": "BIBREF9" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "We then attempted manual annotation. One inexperienced investor and one long-term experienced investor were asked to label a randomly generated subset of the Pre-COVID dataset. They were instructed to label tweets as buy, sell, or hold based on the tweet content and BTC price percentage fluctuation from the previous day (details in A.2). calculated the cosine similarity of each tweet to these three group representations. We selected the match between a tweet and group with the highest cosine similarity to be used as a feature for that tweet. More concretely, each tweet is compared to the DistilBERT representation of the buy, sell, and hold concatenated tweet groups and the highest similarity group is chosen to be used as a feature.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Twitter Data Collection", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Model. From an NLP perspective, frames represent latent abstractions of a discussion and are not equivalent to topics. We hypothesized that how a topic is discussed, or framed, could be identified in an unsupervised manner by analyzing how the tweet content clusters together. In order to extract the clusters which represent such frames, we tried two modeling approaches. First, we used a basic k-means clustering. Second, we implemented the 5 Our task is classification for future use in downstream applications. Thus we do not perform regression or time-series analysis in this paper, but will use time-series in future work. 6 DistilBERT had a 0.6% better performance than BERT. ", |
| "cite_spans": [ |
| { |
| "start": 629, |
| "end": 630, |
| "text": "6", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discourse Framing Clustering", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "We We initially experimented with 32 clusters because 32 is the default number of features that get compressed by the autoencoder. However, we observed that several clusters had similar, overlapping themes and keywords. Therefore, we conducted the rest of our experiments with 10 clusters. Figure 2 shows the number of tweets that fall into each of the 10 initial clusters for each modeling approach. Reduced Dimensions Using SVD. SVD is used to reduce the clusters (0 to 9) to two dimensions to better visualize the frame groupings.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 290, |
| "end": 298, |
| "text": "Figure 2", |
| "ref_id": "FIGREF3" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Discourse Framing Prediction", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "Bitcoin halving 7 , the second concerns trading and investing cryptocurrency, and the third discussed", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discourse Framing Prediction", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "\u2022 Trading Frame: Does the tweet discuss how TOPIC TOP WORDS KNOWLEDGE know, bitcoin, time, blockchain, market, world, buy, change, people, point, today BUSINESS year, thank, start, problem, business, write, stop, plan, risk, reason, check SUPPORT make, think, work, want, day, people, need, use, year, week, support, happen, read HOLD look, price, money, try, build, econ, think, end, tell, idea, people, term, win, hold ", |
| "cite_spans": [ |
| { |
| "start": 60, |
| "end": 420, |
| "text": "KNOWLEDGE know, bitcoin, time, blockchain, market, world, buy, change, people, point, today BUSINESS year, thank, start, problem, business, write, stop, plan, risk, reason, check SUPPORT make, think, work, want, day, people, need, use, year, week, support, happen, read HOLD look, price, money, try, build, econ, think, end, tell, idea, people, term, win, hold", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discourse Framing Prediction", |
| "sec_num": "5.2" |
| }, |
| { |
| "text": "In this section, we explore how cryptocurrency frames change over time and their correlation with cryptocurrency day trading behavior. Section 6.1", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Qualitative Results", |
| "sec_num": "6" |
| }, |
| { |
| "text": "shows the effects of the pandemic on day trading discussions and behaviors. Section 6.2 discusses how day trading behaviors are framed. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Qualitative Results", |
| "sec_num": "6" |
| }, |
| { |
| "text": "Prior to the pandemic, Table 5 shows that the cryp-526 tocurrency tweets were framed in terms of aspects 527 important to cryptocurrency itself, i.e., trading ac-528 tions, applications or uses, and long term store 529 value. Table 6 shows that once the pandemic was TRADING price, bitcoin, usd, market, trading, value, action CRYPTO APPLICATION blockchain, btc, business, use, tech, crypto CRYPTO STORE VALUE bitcoin, people, need, want, use, market, value, years POLITICS world, man, president, america, china, work, government, time ing to take place. The past two times that halving cal frames during a Buy movement is the increase ", |
| "cite_spans": [ |
| { |
| "start": 267, |
| "end": 535, |
| "text": "TRADING price, bitcoin, usd, market, trading, value, action CRYPTO APPLICATION blockchain, btc, business, use, tech, crypto CRYPTO STORE VALUE bitcoin, people, need, want, use, market, value, years POLITICS world, man, president, america, china, work, government, time", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 23, |
| "end": 30, |
| "text": "Table 5", |
| "ref_id": "TABREF8" |
| }, |
| { |
| "start": 226, |
| "end": 233, |
| "text": "Table 6", |
| "ref_id": "TABREF9" |
| } |
| ], |
| "eq_spans": [], |
| "section": "525", |
| "sec_num": null |
| }, |
| { |
| "text": "We have presented a dual modeling pipeline to un- ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "7" |
| }, |
| { |
| "text": "We thank Zachary Yarost and Kasper Standio for annotating the dataset, and the anonymous reviewers for their thoughtful comments and suggestions. tweet volumes and sentiment analysis. In SMU Data groups in arabic online discussions. In Proc. of ACL. structure for person-to-person sentiment analysis.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgments", |
| "sec_num": "8" |
| }, |
| { |
| "text": "INFL. MEDIA CO. 10, 000 \u2212 99, 999 45 -100, 000 \u2212 499, 999 24 5 500, \u2212 999, 2 1 \u2265 1, 000, -- investor that has been investing and following the stock market for the past 5 years, and in the past 2 years has been investing in cryptocurrencies. Annotators were given a randomly selected subset of the Pre-COVID Dataset to label for supervised experiments for the day trading actions prediction.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "QUANTITY", |
| "sec_num": null |
| }, |
| { |
| "text": "The reduced dataset for manual annotation has 798 unique tweets, covering approximately 1% of the total dataset. There are 114 different days represented, with 7 distinct tweets per day. Table 8 shows the prediction accuracy for each label (buy, sell, or hold) for the three different models when using all features. We have also performed ablation studies for the features, which can be included with the final draft of the paper.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 187, |
| "end": 194, |
| "text": "Table 8", |
| "ref_id": "TABREF6" |
| } |
| ], |
| "eq_spans": [], |
| "section": "QUANTITY", |
| "sec_num": null |
| }, |
| { |
| "text": "The combination of NLP and financial applications has been gaining interest in recent years. There have been three recent somewhat related publications working in an economics or financial domain.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "A.4 Somewhat Related Work", |
| "sec_num": null |
| }, |
| { |
| "text": "However, these papers are not directly related to this work and due to page constraints, we have moved them to the Appendix for now. Azzi et al. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "A.4 Somewhat Related Work", |
| "sec_num": null |
| }, |
| { |
| "text": "https://coinmarketcap.com/charts/ 2 https://finance.yahoo.com/quote/AMZN", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://coinmarketcap.com/", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Model. For the day trading prediction model, we", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "About every 4 years or so, the amount of BTC that can be mined (mining capacity) decreases by half.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Junyuan Xie, Ross Girshick, and Ali Farhadi. 2015.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "and Juan 4th Workshop on Representation Learning for NLP (RepL4NLP-2019)", |
| "authors": [ |
| { |
| "first": "Jethin", |
| "middle": [], |
| "last": "Abraham", |
| "suffix": "" |
| }, |
| { |
| "first": "Daniel", |
| "middle": [], |
| "last": "Higdon", |
| "suffix": "" |
| }, |
| { |
| "first": "John", |
| "middle": [], |
| "last": "Nelson", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "194--199", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jethin Abraham, Daniel Higdon, John Nelson, and Juan 4th Workshop on Representation Learning for NLP (RepL4NLP-2019), pages 194-199, Florence, Italy. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "The arab spring| overthrowing the protest paradigm? how the new york times, global voices and twitter covered the egyptian revolution", |
| "authors": [ |
| { |
| "first": "Summer", |
| "middle": [], |
| "last": "Harlow", |
| "suffix": "" |
| }, |
| { |
| "first": "Thomas", |
| "middle": [], |
| "last": "Johnson", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "International Journal of Communication", |
| "volume": "5", |
| "issue": "0", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Summer Harlow and Thomas Johnson. 2011. The arab spring| overthrowing the protest paradigm? how the new york times, global voices and twitter covered the egyptian revolution. International Journal of Communication, 5(0).", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Why are you taking this stance? identifying and classifying reasons in ideological debates", |
| "authors": [ |
| { |
| "first": "Saidul", |
| "middle": [], |
| "last": "Kazi", |
| "suffix": "" |
| }, |
| { |
| "first": "Vincent", |
| "middle": [], |
| "last": "Hasan", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Ng", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proc. of EMNLP", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kazi Saidul Hasan and Vincent Ng. 2014. Why are you taking this stance? identifying and classifying reasons in ideological debates. In Proc. of EMNLP.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Forecasting price of cryptocurrencies using tweets sentiment analysis", |
| "authors": [ |
| { |
| "first": "Arti", |
| "middle": [], |
| "last": "Jain", |
| "suffix": "" |
| }, |
| { |
| "first": "Shashank", |
| "middle": [], |
| "last": "Tripathi", |
| "suffix": "" |
| }, |
| { |
| "first": "Harsh", |
| "middle": [], |
| "last": "Dhardwivedi", |
| "suffix": "" |
| }, |
| { |
| "first": "Pranav", |
| "middle": [], |
| "last": "Saxena", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Arti Jain, Shashank Tripathi, Harsh Dhardwivedi, and Pranav Saxena. 2018. Forecasting price of cryp- tocurrencies using tweets sentiment analysis.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Polarized frames on \"climate change\" and \"global warming\" across countries and states: Evidence from twitter big data", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [ |
| "Mo" |
| ], |
| "last": "Jang", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Hart", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Global Environmental Change", |
| "volume": "32", |
| "issue": "", |
| "pages": "11--17", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Mo Jang and P. Sol Hart. 2015. Polarized frames on \"climate change\" and \"global warming\" across countries and states: Evidence from twitter big data. Global Environmental Change, 32:11-17.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Latent dirichlet allocation (lda) and topic modeling: models, applications, a survey", |
| "authors": [ |
| { |
| "first": "Hamed", |
| "middle": [], |
| "last": "Jelodar", |
| "suffix": "" |
| }, |
| { |
| "first": "Yongli", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Chi", |
| "middle": [], |
| "last": "Yuan", |
| "suffix": "" |
| }, |
| { |
| "first": "Xia", |
| "middle": [], |
| "last": "Feng", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiahui", |
| "middle": [], |
| "last": "Jiang", |
| "suffix": "" |
| }, |
| { |
| "first": "Yanchao", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Liang", |
| "middle": [], |
| "last": "Zhao", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Multimedia Tools and Applications", |
| "volume": "78", |
| "issue": "11", |
| "pages": "15169--15211", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hamed Jelodar, Yongli Wang, Chi Yuan, Xia Feng, Xi- ahui Jiang, Yanchao Li, and Liang Zhao. 2019. La- tent dirichlet allocation (lda) and topic modeling: models, applications, a survey. Multimedia Tools and Applications, 78(11):15169-15211.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Leveraging behavioral and social information for weakly supervised collective classification of political discourse on twitter", |
| "authors": [ |
| { |
| "first": "Kristen", |
| "middle": [], |
| "last": "Johnson", |
| "suffix": "" |
| }, |
| { |
| "first": "Di", |
| "middle": [], |
| "last": "Jin", |
| "suffix": "" |
| }, |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Goldwasser", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proc. of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kristen Johnson, Di Jin, and Dan Goldwasser. 2017. Leveraging behavioral and social information for weakly supervised collective classification of polit- ical discourse on twitter. In Proc. of ACL.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Modeling financial analysts' decision making via the pragmatics and semantics of earnings calls", |
| "authors": [ |
| { |
| "first": "Katherine", |
| "middle": [], |
| "last": "Keith", |
| "suffix": "" |
| }, |
| { |
| "first": "Amanda", |
| "middle": [], |
| "last": "Stent", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "493--503", |
| "other_ids": { |
| "DOI": [ |
| "10.18653/v1/P19-1047" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Katherine Keith and Amanda Stent. 2019. Modeling financial analysts' decision making via the pragmat- ics and semantics of earnings calls. In Proceed- ings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 493-503, Flo- rence, Italy. Association for Computational Linguis- tics.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Twitter sentiment analysis: The good the bad and the omg! In", |
| "authors": [ |
| { |
| "first": "Efthymios", |
| "middle": [], |
| "last": "Kouloumpis", |
| "suffix": "" |
| }, |
| { |
| "first": "Theresa", |
| "middle": [], |
| "last": "Wilson", |
| "suffix": "" |
| }, |
| { |
| "first": "Johanna", |
| "middle": [], |
| "last": "Moore", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Proc. of AAAI Conference on Weblogs and Social Media", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Efthymios Kouloumpis, Theresa Wilson, and Johanna Moore. 2011. Twitter sentiment analysis: The good the bad and the omg! In Proc. of AAAI Conference on Weblogs and Social Media.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Day trading and stock price volatility", |
| "authors": [ |
| { |
| "first": "Petri", |
| "middle": [], |
| "last": "Kyr\u00f6l\u00e4inen", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Journal of Economics and Finance", |
| "volume": "32", |
| "issue": "", |
| "pages": "75--89", |
| "other_ids": { |
| "DOI": [ |
| "10.1007/s12197-007-9006-2" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Petri Kyr\u00f6l\u00e4inen. 2008. Day trading and stock price volatility. Journal of Economics and Finance, 32:75-89.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Social information with graph convolutional networks for political perspective detection in news media", |
| "authors": [ |
| { |
| "first": "Chang", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Goldwasser", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Association for Computational Linguistics (ACL)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Chang Li and Dan Goldwasser. 2019. Social informa- tion with graph convolutional networks for political perspective detection in news media. In Association for Computational Linguistics (ACL).", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Sentiment-based prediction of alternative cryptocurrency price fluctuations using gradient boosting tree model", |
| "authors": [], |
| "year": null, |
| "venue": "Frontiers in Physics", |
| "volume": "7", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sentiment-based prediction of alternative cryptocur- rency price fluctuations using gradient boosting tree model. Frontiers in Physics, 7:98.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Truth about crypto price correlation: How closely does eth follow btc?", |
| "authors": [ |
| { |
| "first": "Julia", |
| "middle": [], |
| "last": "Magas", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Julia Magas. 2020. Truth about crypto price correla- tion: How closely does eth follow btc?", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Networked gatekeeping and networked framing on #egypt", |
| "authors": [ |
| { |
| "first": "Sharon", |
| "middle": [], |
| "last": "Meraz", |
| "suffix": "" |
| }, |
| { |
| "first": "Zizi", |
| "middle": [], |
| "last": "Papacharissi", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "The International Journal of Press/Politics", |
| "volume": "18", |
| "issue": "2", |
| "pages": "138--166", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sharon Meraz and Zizi Papacharissi. 2013. Networked gatekeeping and networked framing on #egypt. The International Journal of Press/Politics, 18(2):138- 166.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "2019. I read crypto twitter for hours daily -here are the accounts that really matter", |
| "authors": [ |
| { |
| "first": "Lesa", |
| "middle": [], |
| "last": "Mone", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "ConsenSys Blog", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lesa Mone. 2019. I read crypto twitter for hours daily -here are the accounts that really matter. In ConsenSys Blog.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Bitcoin: A peer-to-peer electronic cash system", |
| "authors": [ |
| { |
| "first": "Satoshi", |
| "middle": [], |
| "last": "Nakamoto", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Satoshi Nakamoto. 2008. Bitcoin: A peer-to-peer elec- tronic cash system.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Number of americans owning crypto doubled in 2019: Finder", |
| "authors": [ |
| { |
| "first": "Helen", |
| "middle": [], |
| "last": "Partz", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Coin Telegraph", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Helen Partz. 2019. Number of americans owning crypto doubled in 2019: Finder. In Coin Telegraph.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Analyzing stock market movements using twitter sentiment analysis", |
| "authors": [ |
| { |
| "first": "Tushar", |
| "middle": [], |
| "last": "Rao", |
| "suffix": "" |
| }, |
| { |
| "first": "Saket", |
| "middle": [], |
| "last": "Srivastava", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proc. of International Conference on Advances in Social Networks Analysis and Mining", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tushar Rao and Saket Srivastava. 2012. Analyzing stock market movements using twitter sentiment analysis. In Proc. of International Conference on Advances in Social Networks Analysis and Mining.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Why bitcoin has a volatile value", |
| "authors": [ |
| { |
| "first": "Nathan", |
| "middle": [], |
| "last": "Reiff", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Nathan Reiff. 2020. Why bitcoin has a volatile value.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Unsupervised modeling of twitter conversations", |
| "authors": [ |
| { |
| "first": "Alan", |
| "middle": [], |
| "last": "Ritter", |
| "suffix": "" |
| }, |
| { |
| "first": "Colin", |
| "middle": [], |
| "last": "Cherry", |
| "suffix": "" |
| }, |
| { |
| "first": "Bill", |
| "middle": [], |
| "last": "Dolan", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Proc. of NAACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Alan Ritter, Colin Cherry, and Bill Dolan. 2010. Unsu- pervised modeling of twitter conversations. In Proc. of NAACL.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter", |
| "authors": [ |
| { |
| "first": "Victor", |
| "middle": [], |
| "last": "Sanh", |
| "suffix": "" |
| }, |
| { |
| "first": "Lysandre", |
| "middle": [], |
| "last": "Debut", |
| "suffix": "" |
| }, |
| { |
| "first": "Julien", |
| "middle": [], |
| "last": "Chaumond", |
| "suffix": "" |
| }, |
| { |
| "first": "Thomas", |
| "middle": [], |
| "last": "Wolf", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter.", |
| "links": null |
| }, |
| "BIBREF23": { |
| "ref_id": "b23", |
| "title": "Multimodal Multi-Task Financial Risk Forecasting", |
| "authors": [ |
| { |
| "first": "Ramit", |
| "middle": [], |
| "last": "Sawhney", |
| "suffix": "" |
| }, |
| { |
| "first": "Puneet", |
| "middle": [], |
| "last": "Mathur", |
| "suffix": "" |
| }, |
| { |
| "first": "Ayush", |
| "middle": [], |
| "last": "Mangal", |
| "suffix": "" |
| }, |
| { |
| "first": "Piyush", |
| "middle": [], |
| "last": "Khanna", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Rajiv Ratn", |
| "suffix": "" |
| }, |
| { |
| "first": "Roger", |
| "middle": [], |
| "last": "Shah", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Zimmermann", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "456--465", |
| "other_ids": { |
| "DOI": [ |
| "10.1145/3394171.3413752" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ramit Sawhney, Puneet Mathur, Ayush Mangal, Piyush Khanna, Rajiv Ratn Shah, and Roger Zimmermann. 2020. Multimodal Multi-Task Financial Risk Fore- casting, page 456-465. Association for Computing Machinery, New York, NY, USA.", |
| "links": null |
| }, |
| "BIBREF24": { |
| "ref_id": "b24", |
| "title": "Exploiting topic based twitter sentiment for stock prediction", |
| "authors": [ |
| { |
| "first": "Jianfeng", |
| "middle": [], |
| "last": "Si", |
| "suffix": "" |
| }, |
| { |
| "first": "Arjun", |
| "middle": [], |
| "last": "Mukherjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Bing", |
| "middle": [], |
| "last": "Liu", |
| "suffix": "" |
| }, |
| { |
| "first": "Qing", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Huayi", |
| "middle": [], |
| "last": "Li", |
| "suffix": "" |
| }, |
| { |
| "first": "Xiaotie", |
| "middle": [], |
| "last": "Deng", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Proc. of 51st Annual Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jianfeng Si, Arjun Mukherjee, Bing Liu, Qing Li, Huayi Li, and Xiaotie Deng. 2013. Exploiting topic based twitter sentiment for stock prediction. In Proc. of 51st Annual Meeting of the Association for Com- putational Linguistics.", |
| "links": null |
| }, |
| "BIBREF25": { |
| "ref_id": "b25", |
| "title": "Joint models of disagreement and stance in online debate", |
| "authors": [ |
| { |
| "first": "Dhanya", |
| "middle": [], |
| "last": "Sridhar", |
| "suffix": "" |
| }, |
| { |
| "first": "James", |
| "middle": [], |
| "last": "Foulds", |
| "suffix": "" |
| }, |
| { |
| "first": "Bert", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| }, |
| { |
| "first": "Lise", |
| "middle": [], |
| "last": "Getoor", |
| "suffix": "" |
| }, |
| { |
| "first": "Marilyn", |
| "middle": [], |
| "last": "Walker", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proc. of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dhanya Sridhar, James Foulds, Bert Huang, Lise Getoor, and Marilyn Walker. 2015. Joint models of disagreement and stance in online debate. In Proc. of ACL.", |
| "links": null |
| }, |
| "BIBREF26": { |
| "ref_id": "b26", |
| "title": "A frame of mind: Using statistical models for detection of framing and agenda setting campaigns", |
| "authors": [ |
| { |
| "first": "Oren", |
| "middle": [], |
| "last": "Tsur", |
| "suffix": "" |
| }, |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Calacci", |
| "suffix": "" |
| }, |
| { |
| "first": "David", |
| "middle": [], |
| "last": "Lazer", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Proc. of ACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Oren Tsur, Dan Calacci, and David Lazer. 2015. A frame of mind: Using statistical models for detection of framing and agenda setting campaigns. In Proc. of ACL.", |
| "links": null |
| }, |
| "BIBREF27": { |
| "ref_id": "b27", |
| "title": "The bitcoin volatility index price and more", |
| "authors": [ |
| { |
| "first": "Austin", |
| "middle": [], |
| "last": "Tuwiner", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Austin Tuwiner. 2020. The bitcoin volatility index price and more.", |
| "links": null |
| }, |
| "BIBREF28": { |
| "ref_id": "b28", |
| "title": "How traders can use twitter to anticipate bitcoin price moves", |
| "authors": [ |
| { |
| "first": "Tiago", |
| "middle": [], |
| "last": "Vidal", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tiago Vidal. 2020. How traders can use twitter to an- ticipate bitcoin price moves, volume.", |
| "links": null |
| }, |
| "BIBREF29": { |
| "ref_id": "b29", |
| "title": "An empirical analysis of data requirements for financial forecasting with neural networks", |
| "authors": [ |
| { |
| "first": "Steven", |
| "middle": [], |
| "last": "Walczak", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "Journal of management information systems", |
| "volume": "17", |
| "issue": "4", |
| "pages": "203--222", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Steven Walczak. 2001. An empirical analysis of data requirements for financial forecasting with neural networks. Journal of management information sys- tems, 17(4):203-222.", |
| "links": null |
| }, |
| "BIBREF30": { |
| "ref_id": "b30", |
| "title": "Stance classification using dialogic properties of persuasion", |
| "authors": [ |
| { |
| "first": "Marilyn", |
| "middle": [ |
| "A" |
| ], |
| "last": "Walker", |
| "suffix": "" |
| }, |
| { |
| "first": "Pranav", |
| "middle": [], |
| "last": "Anand", |
| "suffix": "" |
| }, |
| { |
| "first": "Robert", |
| "middle": [], |
| "last": "Abbott", |
| "suffix": "" |
| }, |
| { |
| "first": "Ricky", |
| "middle": [], |
| "last": "Grant", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proc. of NAACL", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marilyn A. Walker, Pranav Anand, Robert Abbott, and Ricky Grant. 2012. Stance classification using dia- logic properties of persuasion. In Proc. of NAACL.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "uris": null, |
| "num": null, |
| "text": "Cryptocurrency Day Trading Prediction and Framing Discovery Pipeline.", |
| "type_str": "figure" |
| }, |
| "FIGREF1": { |
| "uris": null, |
| "num": null, |
| "text": "The majority of tweet activity comes from influencer accounts that have between 10,000 and 499,999 followers. There are fewer media accounts, however, these have a much broader reach, e.g.,@nytimes potentially reaches 48.2 million people. More details of the number of tweets collected per account type are presented in Appendix A.1. Using the same accounts, we then collected addi-tional cryptocurrency tweets which occurred during 219 the COVID-19 pandemic time frame: from March 220 2020 until June 2020. These tweets comprise our 221 COVID Dataset.", |
| "type_str": "figure" |
| }, |
| "FIGREF2": { |
| "uris": null, |
| "num": null, |
| "text": "(a) Using k-means.(b) Using DEC (Deep Clustering).", |
| "type_str": "figure" |
| }, |
| "FIGREF3": { |
| "uris": null, |
| "num": null, |
| "text": "Number of Tweets Per Cluster. Both figures show the number of tweets per cluster using initial clusters and BOW features for the Pre-COVID dataset.", |
| "type_str": "figure" |
| }, |
| "FIGREF4": { |
| "uris": null, |
| "num": null, |
| "text": "conducted unsupervised clustering experiments using (1) a basic k-means clustering and (2) deep clustering with autoenconders (DEC) as described in Section 4.2. The encoder outputs were used as inputs to the deep clustering layer, and similar to Hadifar et al., the k-means center clusters were used as initial weights for the deep clustering model. The tweets were randomly shuffled for training. The autoencoder ran for 100 epochs, achieving an accuracy of 99.99% with both training and validation loss on the order of 5.5453e-04 and without overfitting.", |
| "type_str": "figure" |
| }, |
| "FIGREF5": { |
| "uris": null, |
| "num": null, |
| "text": "2a shows six clusters identified in our Pre-COVID Dataset by k-means clustering. Using Singular Value Decomposition (SVD) (Figure 3a) and an analysis of the most frequent words appearing in each cluster, we were able to extract three main clusters: the first cluster included tweets discussing (a) Using k-means. (b) Using DEC (Deep Clustering).", |
| "type_str": "figure" |
| }, |
| "FIGREF6": { |
| "uris": null, |
| "num": null, |
| "text": "Pre-COVID Dataset Cluster Visualization on", |
| "type_str": "figure" |
| }, |
| "FIGREF8": { |
| "uris": null, |
| "num": null, |
| "text": "Predicted Frames and Investment Actions. Each figure shows the quantity of tweets using a certain frame (separated by a grey line) associated with each investment movement action: buy, sell, or hold.6.1 Frames Before and During the Pandemic522 Tables 5 and 6 show the most frequent words ap-523 pearing in each of the four clusters extracted from 524 the Pre-COVID or COVID Dataset, respectively.", |
| "type_str": "figure" |
| }, |
| "FIGREF10": { |
| "uris": null, |
| "num": null, |
| "text": "One interesting event captured by the Trading 537 frame in the COVID-19 Dataset was the BTC halv-538 ing event on May 11, 2020. This halving marks the 539 first quarter of the year as a historical event in the 540 cryptocurrency world because this is the third halv-Frame Most Frequent Words CRYPTO", |
| "type_str": "figure" |
| }, |
| "FIGREF11": { |
| "uris": null, |
| "num": null, |
| "text": "during an indicated Buy move-570 ment. However, the opposite occurs, i.e., all frames 571 increase, when the indicated movement is to Sell.", |
| "type_str": "figure" |
| }, |
| "FIGREF12": { |
| "uris": null, |
| "num": null, |
| "text": "5728 https://news.coinsquare.com/government/governmentinstability-bitcoin/; https://www.un.org/africarenewal/magazine/april-2018-july-2018/africa-could-be-next-frontier-cryptocurrency Regarding both Trading and Application frames, it makes sense to purchase cryptocurrency when nobody is talking about it, and sell it when the interest in those topics rises. The COVID frame having a lower frequency during a Buy movement could indicate that investors feel less threatened by the market instability introduced by the pandemic, which is the opposite of the general sentiment of investors dealing with physical stock exchange markets.", |
| "type_str": "figure" |
| }, |
| "FIGREF13": { |
| "uris": null, |
| "num": null, |
| "text": "derstand how the way influential people and news sources frame cryptocurrency discussions on Twitter affects cryptocurrency day trading. Using classic NLP techniques and cosine similarity between the DistilBERT representations of tweet features and cryptocurrency tweets, we provide a day trading prediction model that is capable of distinguishing between day trading actions such as buy, sell, or hold. Using our features and modeling approach we are able to achieve an accuracy of 88.78% over a 49.72% traditional baseline. Furthermore, we are first to present an unsupervised deep clustering approach to reveal the latent frames used to discuss these day trading behaviors. Our work shows interesting relationships between investment actions and how cryptocurrency discussions are framed on Twitter, as well as how these framing patterns change in response to a pandemic.", |
| "type_str": "figure" |
| }, |
| "FIGREF14": { |
| "uris": null, |
| "num": null, |
| "text": "of experience in both invest-798 ing and trading stocks and cryptocurrencies. One 799 of the annotators was an inexperienced investor, 800 who has never bought or sold cryptocurrencies or 801 stocks. The second annotator is an experienced 802", |
| "type_str": "figure" |
| }, |
| "FIGREF15": { |
| "uris": null, |
| "num": null, |
| "text": "report shared task findings for sentence boundary detection of noisy financial PDFs in the First Workshop on Financial Technology and Natural Language Processing (FinNLP). Keith and Stent compare financial analysts' decision making with fiscal quarter earning calls. Finally, Sawhney et al. uses a multimodal text and audio attention model to predict stock market prices.", |
| "type_str": "figure" |
| }, |
| "TABREF0": { |
| "text": "summarizes the amount of 222 unique tweets per account type in the two portions", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>223</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF1": { |
| "text": "Quantity of Unique Tweets Per User Account Type.", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td colspan=\"2\">domain knowledge, yet also susceptible to differ-</td></tr><tr><td colspan=\"2\">ent investing strategies and conflicting knowledge</td></tr><tr><td colspan=\"2\">from Twitter discussions.</td></tr><tr><td colspan=\"2\">Therefore, we used the price information in the</td></tr><tr><td colspan=\"2\">BTC Historical Price Dataset (Section 3.2) to de-</td></tr><tr><td colspan=\"2\">fine a momentum metric that represents the fluctua-</td></tr><tr><td colspan=\"2\">tion of cryptocurrency costs on a given day:</td></tr><tr><td>momentum =</td><td>P rice</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF2": { |
| "text": "", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>reports the results of two different anno-</td></tr><tr><td>tation approaches, where the true labels are those</td></tr><tr><td>generated by the momentum equation (Eqn. 1).</td></tr><tr><td>First, annotators were asked to label the tweets</td></tr><tr><td>based on their content and give an overall label</td></tr><tr><td>for that particular day based on all their individual</td></tr><tr><td>tweet annotations (shown in the OVERALL DAY</td></tr><tr><td>columns of Table 2). Both annotators performed</td></tr><tr><td>significantly below random guessing, e.g., where</td></tr><tr><td>the expected label was sell.</td></tr><tr><td>Second, they were asked to give another over-</td></tr><tr><td>all annotation for a particular day with the addi-</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF3": { |
| "text": "Annotation Precision Experiments. tional information about the BTC price percentage 309 change from the previous day (shown in the TWEET 310 + PRICE columns). This required annotators to take into consideration the price movement from the experimented with a combination of features, mod-els, and a balanced dataset where there is an equal number of sell, buy, and hold labels. Naive Bayeswith Bag-of-Words (BOW) features was used for the baseline model. We then tested Random Forest, RNN, and LSTM models that resulted in final accuracies above 85%. We conclude in Section 5 that the best performing model for this task 5 is the RNN with three layers.", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>Features. Social network related features are ex-</td></tr><tr><td>tracted directly from the meta-information of the</td></tr><tr><td>cryptocurrency tweets. This includes the number</td></tr><tr><td>of retweets and the number of replies. In our ex-</td></tr><tr><td>periments, we found that the number of retweets</td></tr><tr><td>provided some information gain when weighting</td></tr><tr><td>the tweet feature representation. The type of user</td></tr><tr><td>account, either influencer, media, or company, that</td></tr><tr><td>posted the tweet is also used as a feature.</td></tr><tr><td>In addition to social features, we also used fea-</td></tr><tr><td>tures directly related to the language of the tweet.</td></tr><tr><td>First, we implemented an LDA topic model (Jelo-</td></tr><tr><td>dar et al., 2019) and used the presence of a top 10</td></tr><tr><td>topic in a given tweet as a feature. Next, the tweets</td></tr><tr><td>were transformed into 768 language features using</td></tr><tr><td>DistilBERT. 6 All of the tweets were concatenated</td></tr><tr><td>according to their momentum label and for each</td></tr><tr><td>group (buy, sell, or hold), DistilBERT was used</td></tr><tr><td>to extract high-quality language features to repre-</td></tr><tr><td>sent each of the three tweet groups. Finally, we</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF5": { |
| "text": "Experimental Results. The columns represent the accuracy of each model when using either a bag-ofwords (BOW) or all features of Section 4.1 combined with a DistilBERT(Sanh et al., 2019) representation of the tweets as features.", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>unsupervised Deep Embedded Clustering (DEC)</td><td>389</td></tr><tr><td>approach of Xie et al.; Hadifar et al., which com-</td><td>390</td></tr><tr><td>bines both an autoencoder and k-means clustering</td><td>391</td></tr><tr><td>to achieve a more precise separation. DEC simulta-</td><td/></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF6": { |
| "text": "", |
| "html": null, |
| "num": null, |
| "content": "<table/>", |
| "type_str": "table" |
| }, |
| "TABREF7": { |
| "text": "Pre-COVID Dataset Top 4 LDA Topics and Most Frequent Keywords.", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>or why to buy or sell cryptocurrency?</td></tr><tr><td>\u2022 Application Frame: Does the tweet emphasize</td></tr><tr><td>the uses of cryptocurrency?</td></tr><tr><td>\u2022 Store Value Frame: Does the tweet discuss</td></tr><tr><td>cryptocurrency in terms of long term value?</td></tr><tr><td>\u2022 Political Frame: Does the tweet put a political</td></tr><tr><td>spin on cryptocurrency trading actions?</td></tr><tr><td>The evaluator's manual annotation was compared</td></tr><tr><td>to the actual cluster (or frame) the tweet was as-</td></tr><tr><td>signed to by the DEC model. With this evaluation</td></tr><tr><td>approach, we found the clustering to be 69.23%</td></tr><tr><td>accurate. Given the lack of previous work on cryp-</td></tr><tr><td>tocurrency framing, we compared this result to a</td></tr><tr><td>previous work which found an annotator agreement</td></tr><tr><td>of 73.4% on a tweet dataset labeled for political</td></tr><tr><td>frames (Johnson et al., 2017).</td></tr><tr><td>Next, a chi-square test was performed to verify</td></tr><tr><td>the hypothesis that the frames, represented by clus-</td></tr><tr><td>ters, were independent of each other. In order to</td></tr><tr><td>perform the test, the top word count was collected</td></tr><tr><td>for each cluster, as well as their count in every other</td></tr><tr><td>cluster. The resulting p-value was less than 0.05,</td></tr><tr><td>indicating that the clusters are independent.</td></tr><tr><td>Lastly, to justify that these clusters represent</td></tr><tr><td>how tweets are framed, we also performed an LDA</td></tr><tr><td>topic analysis to ensure that clusters were not find-</td></tr><tr><td>ing topics. Table 4 shows the top 4 LDA topics</td></tr><tr><td>which are different from those extracted for frames</td></tr><tr><td>(more details in Section 6). Topics represent the</td></tr><tr><td>content of the tweet, e.g., the topic Hold represents</td></tr><tr><td>holding cryptocurrency. Frames, however, are fun-</td></tr><tr><td>damentally different and represent how someone</td></tr><tr><td>discusses a topic, e.g., how or why to hold.</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF8": { |
| "text": "Most Frequent Words Per Cluster Prior to COVID-19 (Pre-COVID Dataset).", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>Frame</td><td>Most Frequent Words</td></tr><tr><td>CRYPTO TRADING</td><td>money, crypto, btc, trading, finance, investment, halving</td></tr><tr><td colspan=\"2\">CRYPTO APPLICATION btc, crypto, time, right, know</td></tr><tr><td>SENTIMENT</td><td>like, look, things, dont, good, time, feel</td></tr><tr><td>COVID</td><td>people, coronavirus, covid, pandemic, bitcoin, world, dont</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF9": { |
| "text": "Most Frequent Words Per Cluster During COVID-19 (COVID Dataset).", |
| "html": null, |
| "num": null, |
| "content": "<table/>", |
| "type_str": "table" |
| }, |
| "TABREF10": { |
| "text": "Quantity of Followers Per User Account Type.", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td colspan=\"2\">Each row represents the number of user account types</td></tr><tr><td colspan=\"2\">(columns: influencers, media, company) that have that</td></tr><tr><td colspan=\"2\">quantity of followers who are actively tweeting about</td></tr><tr><td>cryptocurrency.</td><td/></tr><tr><td>MODEL</td><td>BUY SELL HOLD</td></tr><tr><td>NAIVE BAYES</td><td>57 % 62 % 65 %</td></tr><tr><td colspan=\"2\">RANDOM FOREST 82% 87 % 90%</td></tr><tr><td>RNN</td><td>85% 88 % 89%</td></tr></table>", |
| "type_str": "table" |
| }, |
| "TABREF11": { |
| "text": "Label Distribution Results. The columns represent the accuracy of each label based on models using all the features of Section 4.1 combined.From this table, we can see that the majority of", |
| "html": null, |
| "num": null, |
| "content": "<table><tr><td>773</td><td>A Appendix</td></tr><tr><td>774</td><td>A.1 Twitter Data Collection</td></tr><tr><td>775</td><td>In order to determine our Twitter accounts sub-</td></tr><tr><td>776</td><td>set we narrowed it down to accounts that were</td></tr><tr><td>777</td><td>associated with crypto, cryptocurrency, bitcoin</td></tr><tr><td>778</td><td>and blockchain keywords. We only considered</td></tr><tr><td>779</td><td>accounts that had more than 10,000 followers. Fur-</td></tr><tr><td/><td>ther, we segregated the accounts into three distinct</td></tr><tr><td>781</td><td/></tr><tr><td>782</td><td>Table 7 presents the distribution of followers</td></tr><tr><td>783</td><td>for accounts collected from the different types of</td></tr><tr><td>784</td><td>accounts: influencers, media, or company. Col-</td></tr><tr><td>785</td><td>umn one lists the quantity of followers, divided</td></tr><tr><td>786</td><td>into four groups. The remaining columns indicate</td></tr><tr><td>787</td><td>how many of the influencer, media, and company</td></tr><tr><td/><td>accounts have the different number of followers.</td></tr><tr><td>789</td><td/></tr><tr><td>790</td><td>tweet activity comes from influencer accounts that</td></tr><tr><td>791</td><td>have between 10,000 and 499,999 followers. There</td></tr><tr><td>792</td><td>are fewer media accounts, however, these accounts</td></tr><tr><td>793</td><td>have much broader reach. For example, @nytimes</td></tr><tr><td>794</td><td>reaches up to 48.2 million people when tweeting</td></tr><tr><td/><td>about cryptocurrencies.</td></tr></table>", |
| "type_str": "table" |
| } |
| } |
| } |
| } |
