| { |
| "paper_id": "W06-0201", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T04:02:29.042723Z" |
| }, |
| "title": "Development of an automatic trend exploration system using the MuST data collection", |
| "authors": [ |
| { |
| "first": "Masaki", |
| "middle": [], |
| "last": "Murata", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "murata@nict.go.jp" |
| }, |
| { |
| "first": "Qing", |
| "middle": [], |
| "last": "Ma", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Institute of Information and Communications Technology", |
| "location": { |
| "addrLine": "3-5 Hikaridai, Seika-cho, Soraku-gun", |
| "postCode": "619-0289", |
| "settlement": "Kyoto", |
| "country": "Japan" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Toshiyuki", |
| "middle": [], |
| "last": "Kanamaru", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "National Institute of Information and Communications Technology", |
| "location": { |
| "addrLine": "3-5 Hikaridai, Seika-cho, Soraku-gun", |
| "postCode": "619-0289", |
| "settlement": "Kyoto", |
| "country": "Japan" |
| } |
| }, |
| "email": "kanamaru@nict.go.jp" |
| }, |
| { |
| "first": "Hitoshi", |
| "middle": [], |
| "last": "Isahara", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "isahara@nict.go.jp" |
| }, |
| { |
| "first": "Tamotsu", |
| "middle": [], |
| "last": "Shirado", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "shirado@nict.go.jp" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The automatic extraction of trend information from text documents such as newspaper articles would be useful for exploring and examining trends. To enable this, we used data sets provided by a workshop on multimodal summarization for trend information (the MuST Workshop) to construct an automatic trend exploration system. This system first extracts units, temporals, and item expressions from newspaper articles, then it extracts sets of expressions as trend information, and finally it arranges the sets and displays them in graphs. For example, when documents concerning the politics are given, the system extracts \"%\" and \"Cabinet approval rating\" as a unit and an item expression including temporal expressions. It next extracts values related to \"%\". Finally, it makes a graph where temporal expressions are used for the horizontal axis and the value of percentage is shown on the vertical axis. This graph indicates the trend of Cabinet approval rating and is useful for investigating Cabinet approval rating. Graphs are obviously easy to recognize and useful for understanding information described in documents. In experiments, when we judged the extraction of a correct graph as the top output to be correct, the system accuracy was 0.2500 in evaluation A and 0.3334 in evaluation B. (In evaluation A, a graph where 75% or more of the points were correct was judged to be correct; in evaluation B, a graph where 50% or more of the points were correct was judged to be correct.) When we judged the extraction of a correct graph in the top five outputs to be correct, accuracy rose to 0.4167 in evaluation A and 0.6250 in evaluation B. Our system is convenient and effective because it can output a graph that includes trend information at these levels of accuracy when given only a set of documents as input.", |
| "pdf_parse": { |
| "paper_id": "W06-0201", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The automatic extraction of trend information from text documents such as newspaper articles would be useful for exploring and examining trends. To enable this, we used data sets provided by a workshop on multimodal summarization for trend information (the MuST Workshop) to construct an automatic trend exploration system. This system first extracts units, temporals, and item expressions from newspaper articles, then it extracts sets of expressions as trend information, and finally it arranges the sets and displays them in graphs. For example, when documents concerning the politics are given, the system extracts \"%\" and \"Cabinet approval rating\" as a unit and an item expression including temporal expressions. It next extracts values related to \"%\". Finally, it makes a graph where temporal expressions are used for the horizontal axis and the value of percentage is shown on the vertical axis. This graph indicates the trend of Cabinet approval rating and is useful for investigating Cabinet approval rating. Graphs are obviously easy to recognize and useful for understanding information described in documents. In experiments, when we judged the extraction of a correct graph as the top output to be correct, the system accuracy was 0.2500 in evaluation A and 0.3334 in evaluation B. (In evaluation A, a graph where 75% or more of the points were correct was judged to be correct; in evaluation B, a graph where 50% or more of the points were correct was judged to be correct.) When we judged the extraction of a correct graph in the top five outputs to be correct, accuracy rose to 0.4167 in evaluation A and 0.6250 in evaluation B. Our system is convenient and effective because it can output a graph that includes trend information at these levels of accuracy when given only a set of documents as input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "We have studied ways to automatically extract trend information from text documents, such as newspaper articles, because such a capability will be useful for exploring and examining trends. In this work, we used data sets provided by a workshop on multimodal summarization for trend information (the MuST Workshop) to construct an automatic trend exploration system. This system firsts extract units, temporals, and item expressions from newspaper articles, then it extract sets of expressions as trend information, and finally it arranges the sets and displays them in graphs. For example, when documents concerning the politics are given, the system extracts \"%\" and \"Cabinet approval rating\" as a unit and an item expression including temporal expressions. It next extracts values related to \"%\". Finally, it makes a graph where temporal expressions are used for the horizontal axis and the value of percentage is shown on the vertical axis. This graph indicates the trend of Cabinet approval rating and is useful for investigating Cabinet approval rating. Graphs are obviously easy to recognize and useful for understanding information described in documents.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "Kato et al. organized the workshop on multimodal summarization for trend information (the MuST Workshop) (Kato et al., 2005) . In this workshop, participants were given data sets consisting of newspaper documents (editions of the Mainichi newspaper from 1998 and 1999 (Japanese documents)) that included trend information for various domains. In the data, tags for important expressions (e.g. temporals, numerical expressions, and item expressions) were tagged manually. 1 The 20 topics of the data sets (e.g., the 1998 home-run race to break the all-time Major League record, the approval rating for the Japanese Cabinet, and news on typhoons) were provided. Trend information was defined as information regarding the change in a value for a certain item. A change in the number of home runs hit by a certain player or a change in the approval rating for the Cabinet are examples of trend information. In the workshop, participants could freely use the data sets for any study they chose to do.", |
| "cite_spans": [ |
| { |
| "start": 105, |
| "end": 124, |
| "text": "(Kato et al., 2005)", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 471, |
| "end": 472, |
| "text": "1", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The MuST Workshop", |
| "sec_num": "2" |
| }, |
| { |
| "text": "Our automatic trend exploration system consists of the following components.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Structure of the system", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "First, documents related to a certain topic are given to the system, which then extracts important expressions that will be used to extract and merge trend information. The system extracts item units, temporal units, and item expressions as important expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "1." |
| }, |
| { |
| "text": "Here, important expressions are defined as expressions that play important roles in a given document set. Item expressions are defined as expressions that are strongly related to the content of a given document set.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "1." |
| }, |
| { |
| "text": "1a. Component to extract important item units The system extracts item units that will be used to extract and merge trend information. For example, when documents concerning the home-run race are given, \"hon\" or \"gou\" (the Japanese item units for the number of home runs) such as in \"54 hon\" (54th home run) are extracted. 1b. Component to extract important temporal units", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "1." |
| }, |
| { |
| "text": "The system extracts temporal units that will also be used to extract and merge trend information. For example, the system extracts temporal units such as \"nichi\" (day), \"gatsu\" (month), and \"nen\" (year). In Japanese, temporal units are used to express dates, such as in \"2006 nen, 3 gatsu, 27 nichi\" for March 27th, 2006. 1c. Component to extract important item expressions The system extracts item expressions that will also be used to extract and merge trend information. For example, the system extracts expressions that are objects for trend exploration, such as \"McGwire\" and \"Sosa\" as item expressions in the case of documents concerning the home-run race.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "1." |
| }, |
| { |
| "text": "The system identifies the locations in sentences where a temporal unit, an item unit, and an item expression that was extracted by the component to extract important expressions appear in similar sentences and extracts sets of important expressions described by the sentences as a trend information set. The system also extracts numerical values appearing with item units or temporal units, and uses the connection of the numerical values and the item units or temporal units as numerical expressions or temporal expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "For example, in the case of documents concerning the home-run race, the system extracts a set consisting of \"item expression:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "McGwire\", \"temporal expression: 11 day\" (the 11th), and \"numerical expression: 47 gou\" (47th home run) as a trend information set.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "3. Component to extract and display important trend information sets", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "The system gathers the extracted trend information sets and displays them as graphs or by highlighting text displays.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "For example, for documents concerning the home-run race, the system displays as graphs the extracted trend information sets for \"McGwire\" . In these graphs, temporal expressions are used for the horizontal axis and the number of home runs is shown on the vertical axis.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "2." |
| }, |
| { |
| "text": "The system extracts important expressions that will be used to extract trend information sets. Important expressions belong to one of the following categories.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "\u2022 item units", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "\u2022 temporal units \u2022 item expressions", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "We use ChaSen (Matsumoto et al., 1999) , a Japanese morphological analyzer, to extract expressions. Specifically, we use the parts of speeches in the ChaSen outputs to extract the expressions.", |
| "cite_spans": [ |
| { |
| "start": 14, |
| "end": 38, |
| "text": "(Matsumoto et al., 1999)", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The system extracts item units, temporal units, and item expressions by using manually constructed rules using the parts of speeches. The system extracts a sequence of nouns adjacent to numerical values as item units. It then extracts expressions from among the item units which include an expression regarding time or date (e.g., \"year\", \"month\", \"day\", \"hour\", or \"second\") as temporal units. The system extracts a sequence of nouns as item expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The system next extracts important item units, temporal units, and item expressions that play important roles in the target documents.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The following three methods can be used to extract important expressions. The system uses one of them. The system judges that an expression producing a high value from the following equations is an important expression.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "\u2022 Equation for the TF numerical term in Okapi (Robertson et al., 1994 )", |
| "cite_spans": [ |
| { |
| "start": 46, |
| "end": 69, |
| "text": "(Robertson et al., 1994", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "Score = i\u2208Docs T F i T F i + l i \u2206 (1) \u2022 Use of total word frequency Score = i\u2208Docs T F i (2)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "\u2022 Use of total frequency of documents where a word appears", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "Score = i\u2208Docs 1", |
| "eq_num": "(3)" |
| } |
| ], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "In these equations, i is the ID (identification number) of a document, Docs is a set of document IDs, T F i is the occurrence number of an expression in document i, l is the length of document i, and \u2206 is the average length of documents in Docs.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "To extract item expressions, we also applied a method that uses the product of the occurrence number of an expression in document i and the length of the expression as T F i , so that we could extract longer expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract important expressions", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The system identifies the locations in sentences where a temporal unit, an item unit, and an item expression extracted by the component to extract important expressions appears in similar sentences and extracts sets of important expressions described by the sentences as a trend information set. When more than one trend information set appears in a document, the system extracts the one that appears first. This is because important and new things are often described in the beginning of a document in the case of newspaper articles.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract trend information sets", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "The system gathers the extracted trend information sets and displays them in graphs or as highlighted text. In the graphs, temporal expressions are used for the horizontal axis and numerical expressions are used for the vertical axis. The system also displays sentences used to extract trend information sets and highlights important expressions in the sentences.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "The system extracts multiple item units, temporal units, and item expressions (through the component to extract important expressions) and uses these to make all possible combinations of the three kinds of expression. The system extracts trend information sets for each combination and calculates the value of one of the following equations for each combination. The system judges that the combination producing a higher value represents more useful trend information. The following four equations can be used for this purpose, and the system uses one of them.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "\u2022 Method 1 -Use both the frequency of trend information sets and the scores of important expressions", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "M = F req \u00d7 S 1 \u00d7 S 2 \u00d7 S 3 (4)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "\u2022 Method 2 -Use both the frequency of trend information sets and the scores of important expressions", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "M = F req \u00d7 (S 1 \u00d7 S 2 \u00d7 S 3 ) 1 3", |
| "eq_num": "(5)" |
| } |
| ], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "\u2022 Method 3 -Use the frequency of trend information sets", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "M = F req (6)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "\u2022 Method 4 -Use the scores of important expressions", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "M = S 1 \u00d7 S 2 \u00d7 S 3", |
| "eq_num": "(7)" |
| } |
| ], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "In these equations, F req is the number of trend information sets extracted as described in Section 3.3, and S1, S2, and S3 are the values of Score as calculated by the corresponding equation in Section 3.2.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "The system extracts the top five item units, the top five item expressions, and the top three temporal units through the component to extract important expressions and forms all possible combinations of these (75 combinations). The system then calculates the value of the above equations for these 75 combinations and judges that a combination having a larger value represents more useful trend information.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Component to extract and display important trend information sets", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "We describe some examples of the output of our system in Sections 4.1, 4.2, and 4.3, and the results from our system evaluation in Section 4.4. We made experiments using Japanese newspaper articles.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Discussion", |
| "sec_num": "4" |
| }, |
| { |
| "text": "To extract important expressions we applied the equation for the TF numerical term in Okapi and the method using the product of the occurrence number for an expression and the length of the expression as T F i for item expressions. We did experiments using the three document sets for typhoons, the Major Leagues, and political trends. The results are shown in Table 1 .", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 361, |
| "end": 368, |
| "text": "Table 1", |
| "ref_id": "TABREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Extracting important expressions", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "We found that appropriate important expressions were extracted for each domain. For example, in the data set for typhoons, \"typhoon\" was extracted as an important item expression and an item unit \"gou\" (No.), indicating the ID number of each typhoon, was extracted as an important item unit. In the data set for the Major Leagues, the MuST data included documents describing the home-run race between Mark McGwire and Sammy Sosa in 1998. \"McGwire\" and \"Sosa\" were properly extracted among the higher ranks. \"gou\" (No.) and \"hon\" (home run(s)), important item units for the home-run race, were properly extracted. In the data set for political trends, \"naikaku shiji ritsu\" (cabinet approval rating) was properly extracted as an item expression and \"%\" was extracted as an item unit.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracting important expressions", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "We next tested how well our system graphed the trend information obtained from the MuST data sets. We used the same three document sets as in the previous section. As important expressions in the experiments, we used the item unit, the temporal unit, and the item expression with the highest scores (the top ranked ones) which were extracted by the component to extract important expressions using the method described in the previous section. The system made the graphs using the component to extract trend information sets and the component to extract and display important trend information sets. The graphs thus produced are shown in Figs. 1, 2, and 3. (We used Excel to draw these graphs.) Here, we made a temporal axis for each temporal expression. However, we can also For the typhoon data set, gou (No.), nichi (day), and taihuu (typhoon) were respectively extracted as the top ranked item unit, temporal unit, and item expression. The system extracted trend information sets using these, and then made a graph where the temporal expression (day) was used for the horizontal axis and the ID numbers of the typhoons were shown on the vertical axis. The MuST data included data for September and October of 1998 and 1999. Figure 1 is useful for seeing when each typhoon hit Japan during the typhoon season each year. Comparing the 1998 data with that of 1999 reveals that the number of typhoons increased in 1999.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 1228, |
| "end": 1236, |
| "text": "Figure 1", |
| "ref_id": "FIGREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Graphs representing trend information", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "For the Major Leagues data set, gou (No.), nichi (day), and Maguwaia (McGwire) were extracted with the top rank. The system used these to make a graph where the temporal expression (day) was used for the horizontal axis and the cumulative number of home runs hit by McGwire was shown on the vertical axis (Fig. 2) . The MuST data included data beginning in August, 1998. The graph shows some points where the cumulative number of home runs decreased (e.g., September Figure 3 : Trend graph for the political trends data set 4th), which was obviously incorrect. This was because our system wrongly extracted the number of home runs hit by Sosa when this was given close to McGwire's total.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 305, |
| "end": 313, |
| "text": "(Fig. 2)", |
| "ref_id": "FIGREF1" |
| }, |
| { |
| "start": 467, |
| "end": 475, |
| "text": "Figure 3", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Graphs representing trend information", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "In the political trends data set, %, gatsu (month), and naikaku shiji ritsu (cabinet approval rating) were extracted with the top rankings. The system used these to make a graph where the temporal expression (month) was used for the horizontal axis and the Cabinet approval rating (Japanese Cabinet) was shown as a percentage on the vertical axis. The MuST data covered 1998 and 1999. Figure 2 shows the cabinet approval rating of the Obuchi Cabinet. We found that the overall approval rating trend was upwards. Again, there were some errors in the extracted trend information sets. For example, although June was handled correctly, the system wrongly extracted May as a temporal expression from the sentence \"in comparison to the previous investigation in May\".", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 385, |
| "end": 393, |
| "text": "Figure 2", |
| "ref_id": "FIGREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Graphs representing trend information", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "We then tested the sentence extraction and highlighting display with respect to trend information using the MuST data set; in this case, we used the typhoon data set. As important expressions, we used the item unit, the temporal unit, and the item expression extracted with the highest scores (the top ranked ones) by the component to extract important expressions using the method described in the previous section. Gou (No.) , nichi (day), and taihuu (typhoon) were respectively extracted as an item unit, a temporal unit, and an item expression. The system extracted sentences including the three expressions and highlighted these expressions in the sentences. The results are shown in Figure 4 . The first trend information sets to ap- :00 a.m. on the 24th, and after moving to Suo-Nada made another landfall at Ube City in Yamaguchi Prefecture before 9:00 p.m., tracked through the Chugoku district, and then moved into the Japan Sea after 10:00 p.m. . 25, 1999 No. 18 Typhoon No. 18 , which caused significant damage in the Kyushu and Chugoku districts, weakened and made another landfall before moving into the Sea of Okhotsk around 10:00 a.m. on the 25th. pear are underlined twice and the other sets are underlined once. (In the actual system, color is used to make this distinction.) The extracted temporal expressions and numerical expressions are presented in the upper part of the extracted sentence. The graphs shown in the previous section were made by using these temporal expressions and numerical expressions. The extracted sentences plainly described the state of affairs regarding the typhoons and were important sentences. For the research being done on summarization techniques, this can be considered a useful means of extracting important sentences. The extracted sentences typically describe the places affected by each typhoon and whether there was any damage. They contain important descriptions about each typhoon. This confirmed that a simple method of extracting sentences containing an item unit, a temporal unit, and an item expression can be used to extract important sentences.", |
| "cite_spans": [ |
| { |
| "start": 421, |
| "end": 426, |
| "text": "(No.)", |
| "ref_id": null |
| }, |
| { |
| "start": 956, |
| "end": 988, |
| "text": ". 25, 1999 No. 18 Typhoon No. 18", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 689, |
| "end": 697, |
| "text": "Figure 4", |
| "ref_id": "FIGREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Sentence extraction and highlighting display", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "The fourth sentence in the figure includes information on both typhoon no.7 and typhoon no.8. We can see that there is a trend information set other than the extracted trend information set (underlined twice) from the expressions that are underlined once. Since the system sometimes extracts incorrect trend information sets, the highlighting is useful for identifying such sets.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Sept", |
| "sec_num": null |
| }, |
| { |
| "text": "We used a closed data set and an open data set to evaluate our system. The closed data set was the data set provided by the MuST workshop organizer and contained 20 domain document sets. The data sets were separated for each domain.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "We made the open data set based on the MuST data set using newspaper articles (editions of the Mainichi newspaper from 2000 and 2001). We made 24 document sets using information retrieval by term query. We used documents retrieved by term query as the document set of the domain for each query term.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "We used the closed data set to adjust our system and used the open data set to calculate the evaluation scores of our system for evaluation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "We judged whether a document set included the information needed to make trend graphs by consulting the top 30 combinations of three kinds of important expression having the 30 highest values as in the method of Section 3.4. There were 19 documents including such information in the open data. We used these 19 documents for the following evaluation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "In the evaluation, we examined how accurately trend graphs could be output when using the top ranked expressions. The results are shown in Table 2 . The best scores are described using bold fonts for each evaluation score.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 139, |
| "end": 147, |
| "text": "Table 2", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "We used five evaluation scores. MRR is the average of the score where 1/r is given as the score when the rank of the first correct output is r (Murata et al., 2005b) . TP1 is the average of the precision in the first output. TP5 is the average of the precision where the system includes a correct output in the first five outputs. RP is the average of the r-precision and AP is the average of the average precision. (Here, the average means that the evaluation score is calculated for each domain data set and the summation of these scores divided by the number of the domain data sets is the average.) R-precision is the precision of the r outputs where r is the number of correct answers. Average precision is the average of the precision when each correct answer is output (Murata et al., 2000) . The r-precision indicates the precision where the recall and the precision have the same value. The precision is the ratio of correct answers in the system output. The recall is the ratio of correct answers in the system output to the total number of correct answers.", |
| "cite_spans": [ |
| { |
| "start": 143, |
| "end": 165, |
| "text": "(Murata et al., 2005b)", |
| "ref_id": "BIBREF5" |
| }, |
| { |
| "start": 776, |
| "end": 797, |
| "text": "(Murata et al., 2000)", |
| "ref_id": "BIBREF3" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "Methods 1 to 4 in Table 2 are the methods used to extract useful trend information described in Section 3.4. Use of the expression length means the product of the occurrence number for an expression and the length of the expression was used to calculate the score for an important item expression. No use of the expression length means this product was not used and only the occurrence number was used.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 18, |
| "end": 25, |
| "text": "Table 2", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "To calculate the r-precision and average precision, we needed correct answer sets. We made the correct answer sets by manually examining the top 30 outputs for the 24 (= 4 \u00d7 6) methods (the combinations of methods 1 to 4 and the use of Equations 1 to 3 with or without the expression length) and defining the useful trend information among them as the correct answer sets.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "In evaluation A, a graph where 75% or more of the points were correct was judged to be correct. In evaluation B, a graph where 50% or more of the points were correct was judged to be correct. From the experimental results, we found that the method using the total frequency for a word (Equation 2) and the length of an expression was best for calculating the scores of important expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "Using the length of an expression was important. (The way of using the length of an expression was described in the last part of Section 3.2.) For example, when \"Cabinet approval rating\" appears in documents, a method without expression lengths extracts \"rating\". When the system extracts trend information sets using \"rating\", it extracts wrong information related to types of \"rating\" other than \"Cabinet approval rating\". This hinders the extraction of coherent trend information. Thus, it is beneficial to use the length of an expression when extracting important item expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "We also found that method 1 (using both the frequency of the trend information sets and the scores of important expressions) was generally the best.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "When we judged the extraction of a correct graph as the top output in the experiments to be correct, our best system accuracy was 0.3158 in evaluation A and 0.4211 in evaluation B. When we judged the extraction of a correct graph in the top five outputs to be correct, the best accuracy rose to 0.5263 in evaluation A and 0.7895 in evaluation B. In terms of the evaluation scores for the 24 original data sets (these evaluation scores were multiplied by 19/24), when we judged the extraction of a correct graph as the top output in the experiments to be correct, our best system accuracy was 0.3158 in evaluation A and 0.4211 in evaluation B. When we judged the extraction of a correct graph in the top five outputs to be correct, the best accuracy rose to 0.5263 in evaluation A and 0.7895 in evaluation B. Our system is convenient and effective because it can output a graph that includes trend information at these levels of accuracy when given only a set of documents as input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "As shown in Table 2 , the best values for RP (which indicates the precision where the recall and the precision have the same value) and AP were 0.2127 and 0.1705, respectively, in evaluation B.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 12, |
| "end": 19, |
| "text": "Table 2", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "This RP value indicates that our system could extract about one out of five graphs among the correct answers when the recall and the precision had the same value.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Evaluation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "Fujihata et al. (Fujihata et al., 2001 ) developed a system to extract numerical expressions and their related item expressions by using syntactic information and patterns. However, they did not deal with the extraction of important expressions or gather trend information sets. In addition, they did not make a graph from the extracted expressions. Nanba et al. (Nanba et al., 2005) took an approach of judging whether the sentence relationship indicates transition (trend information) or renovation (revision of information) and used the judgment results to extract trend information. They also constructed a system to extract numerical information from input numerical units and make a graph that includes trend information. However, they did not consider ways to extract item numerical units and item expressions automatically.", |
| "cite_spans": [ |
| { |
| "start": 16, |
| "end": 38, |
| "text": "(Fujihata et al., 2001", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 350, |
| "end": 383, |
| "text": "Nanba et al. (Nanba et al., 2005)", |
| "ref_id": "BIBREF6" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related studies", |
| "sec_num": "5" |
| }, |
| { |
| "text": "In contrast to these systems, our system automatically extracts item numerical units and item expressions that each play an important role in a given document set. When a document set for a certain domain is given, our system automatically extracts item numerical units and item expressions, then extracts numerical expressions related to these, and finally makes a graph based on the extracted numerical expressions. When a document set is given, the system automatically makes a graph that includes trend information. Our system also uses an original method of producing more than one graphs and selecting an appropriate graph among them using Methods 1 to 4, which Fujihata et al. and Namba et al. did not use.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related studies", |
| "sec_num": "5" |
| }, |
| { |
| "text": "We have studied the automatic extraction of trend information from text documents such as newspaper articles. Such extraction will be useful for exploring and examining trends. We used data sets provided by a workshop on multimodal summarization for trend information (the MuST Workshop) to construct our automatic trend exploration system. This system first extracts units, temporals, and item expressions from newspaper articles, then it extracts sets of expressions as trend information, and finally it arranges the sets and displays them in graphs.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "6" |
| }, |
| { |
| "text": "In our experiments, when we judged the extraction of a correct graph as the top output to be correct, the system accuracy was 0.2500 in evaluation A and 0.3334 in evaluation B. (In evaluation A, a graph where 75% or more of the points were correct was judged to be correct; in evaluation B, a graph where 50% or more of the points were correct was judged to be correct.) When we judged the extraction of a correct graph in the top five outputs to be correct, we obtained accuracy of 0.4167 in evaluation A and 0.6250 in evaluation B. Our system is convenient and effective because it can output a graph that includes trend information at these levels of accuracy when only a set of documents is provided as input.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "6" |
| }, |
| { |
| "text": "In the future, we plan to continue this line of study and improve our system. We also hope to apply the method of using term frequency in documents to extract trend information as reported by Murata et al. (Murata et al., 2005a) .", |
| "cite_spans": [ |
| { |
| "start": 192, |
| "end": 228, |
| "text": "Murata et al. (Murata et al., 2005a)", |
| "ref_id": "BIBREF4" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "6" |
| }, |
| { |
| "text": "We do not use manually provided tags for important expressions because our system automatically extracts important expressions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Extracting of numerical expressions by constraints and default rules of dependency structure", |
| "authors": [ |
| { |
| "first": "Katsuyuki", |
| "middle": [], |
| "last": "Fujihata", |
| "suffix": "" |
| }, |
| { |
| "first": "Masahiro", |
| "middle": [], |
| "last": "Shiga", |
| "suffix": "" |
| }, |
| { |
| "first": "Tatsunori", |
| "middle": [], |
| "last": "Mori", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Katsuyuki Fujihata, Masahiro Shiga, and Tatsunori Mori. 2001. Extracting of numerical expressions by constraints and default rules of dependency struc- ture. Information Processing Society of Japan, WGNL 145.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "MuST: A workshop on multimodal summarization for trend information", |
| "authors": [ |
| { |
| "first": "Tsuneaki", |
| "middle": [], |
| "last": "Kato", |
| "suffix": "" |
| }, |
| { |
| "first": "Mitsunori", |
| "middle": [], |
| "last": "Matsushita", |
| "suffix": "" |
| }, |
| { |
| "first": "Noriko", |
| "middle": [], |
| "last": "Kando", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proceedings of the Fifth NTCIR Workshop Meeting on Evaluation of Information Access Technologies: Information Retrieval, Question Answering and Cross-Lingual Information Access", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tsuneaki Kato, Mitsunori Matsushita, and Noriko Kando. 2005. MuST: A workshop on multimodal summarization for trend information. Proceedings of the Fifth NTCIR Workshop Meeting on Evaluation of Information Access Technologies: Information Retrieval, Question Answering and Cross-Lingual Information Access.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Japanese morphological analysis system ChaSen version 2", |
| "authors": [ |
| { |
| "first": "Yuji", |
| "middle": [], |
| "last": "Matsumoto", |
| "suffix": "" |
| }, |
| { |
| "first": "Akira", |
| "middle": [], |
| "last": "Kitauchi", |
| "suffix": "" |
| }, |
| { |
| "first": "Tatsuo", |
| "middle": [], |
| "last": "Yamashita", |
| "suffix": "" |
| }, |
| { |
| "first": "Yoshitaka", |
| "middle": [], |
| "last": "Hirano", |
| "suffix": "" |
| }, |
| { |
| "first": "Hiroshi", |
| "middle": [], |
| "last": "Matsuda", |
| "suffix": "" |
| }, |
| { |
| "first": "Masayuki", |
| "middle": [], |
| "last": "Asahara", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Yuji Matsumoto, Akira Kitauchi, Tatsuo Yamashita, Yoshitaka Hirano, Hiroshi Matsuda, and Masayuki Asahara. 1999. Japanese morphological analysis system ChaSen version 2.0 manual 2nd edition.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Japanese probabilistic information retrieval using location and category information", |
| "authors": [ |
| { |
| "first": "Masaki", |
| "middle": [], |
| "last": "Murata", |
| "suffix": "" |
| }, |
| { |
| "first": "Kiyotaka", |
| "middle": [], |
| "last": "Uchimoto", |
| "suffix": "" |
| }, |
| { |
| "first": "Hiromi", |
| "middle": [], |
| "last": "Ozaku", |
| "suffix": "" |
| }, |
| { |
| "first": "Qing", |
| "middle": [], |
| "last": "Ma", |
| "suffix": "" |
| }, |
| { |
| "first": "Masao", |
| "middle": [], |
| "last": "Utiyama", |
| "suffix": "" |
| }, |
| { |
| "first": "Hitoshi", |
| "middle": [], |
| "last": "Isahara", |
| "suffix": "" |
| } |
| ], |
| "year": 2000, |
| "venue": "The Fifth International Workshop on Information Retrieval with Asian Languages", |
| "volume": "", |
| "issue": "", |
| "pages": "81--88", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Masaki Murata, Kiyotaka Uchimoto, Hiromi Ozaku, Qing Ma, Masao Utiyama, and Hitoshi Isahara. 2000. Japanese probabilistic information retrieval using location and category information. The Fifth International Workshop on Information Retrieval with Asian Languages, pages 81-88.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Trend survey on Japanese natural language processing studies over the last decade", |
| "authors": [ |
| { |
| "first": "Masaki", |
| "middle": [], |
| "last": "Murata", |
| "suffix": "" |
| }, |
| { |
| "first": "Koji", |
| "middle": [], |
| "last": "Ichii", |
| "suffix": "" |
| }, |
| { |
| "first": "Qing", |
| "middle": [], |
| "last": "Ma", |
| "suffix": "" |
| }, |
| { |
| "first": "Tamotsu", |
| "middle": [], |
| "last": "Shirado", |
| "suffix": "" |
| }, |
| { |
| "first": "Toshiyuki", |
| "middle": [], |
| "last": "Kanamaru", |
| "suffix": "" |
| }, |
| { |
| "first": "Hitoshi", |
| "middle": [], |
| "last": "Isahara", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "The Second International Joint Conference on Natural Language Processing, Companion Volume to the Proceedings of Conference including Posters/Demos and Tutorial Abstracts", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Masaki Murata, Koji Ichii, Qing Ma, Tamotsu Shirado, Toshiyuki Kanamaru, and Hitoshi Isahara. 2005a. Trend survey on Japanese natural language process- ing studies over the last decade. In The Second In- ternational Joint Conference on Natural Language Processing, Companion Volume to the Proceedings of Conference including Posters/Demos and Tutorial Abstracts.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Use of multiple documents as evidence with decreased adding in a Japanese question-answering system", |
| "authors": [ |
| { |
| "first": "Masaki", |
| "middle": [], |
| "last": "Murata", |
| "suffix": "" |
| }, |
| { |
| "first": "Masao", |
| "middle": [], |
| "last": "Utiyama", |
| "suffix": "" |
| }, |
| { |
| "first": "Hitoshi", |
| "middle": [], |
| "last": "Isahara", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Journal of Natural Language Processing", |
| "volume": "12", |
| "issue": "2", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Masaki Murata, Masao Utiyama, and Hitoshi Isahara. 2005b. Use of multiple documents as evidence with decreased adding in a Japanese question-answering system. Journal of Natural Language Processing, 12(2).", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Extraction and visualization of trend information based on the cross-document structure", |
| "authors": [ |
| { |
| "first": "Hidetsugu", |
| "middle": [], |
| "last": "Nanba", |
| "suffix": "" |
| }, |
| { |
| "first": "Yoshinobu", |
| "middle": [], |
| "last": "Kunimasa", |
| "suffix": "" |
| }, |
| { |
| "first": "Shiho", |
| "middle": [], |
| "last": "Fukushima", |
| "suffix": "" |
| }, |
| { |
| "first": "Teruaki", |
| "middle": [], |
| "last": "Aizawa", |
| "suffix": "" |
| }, |
| { |
| "first": "Manabu", |
| "middle": [], |
| "last": "Okumura", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "WGNL", |
| "volume": "168", |
| "issue": "", |
| "pages": "67--74", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hidetsugu Nanba, Yoshinobu Kunimasa, Shiho Fukushima, Teruaki Aizawa, and Manabu Oku- mura. 2005. Extraction and visualization of trend information based on the cross-document structure. Information Processing Society of Japan, WGNL 168, pages 67-74.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Okapi at TREC-3", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [ |
| "E" |
| ], |
| "last": "Robertson", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Walker", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Jones", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [ |
| "M" |
| ], |
| "last": "Hancock-Beaulieu", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Gatford", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. E. Robertson, S. Walker, S. Jones, M. M. Hancock- Beaulieu, and M. Gatford. 1994. Okapi at TREC-3. In TREC-3.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "text": "Trend graph for the typhoon data set", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "FIGREF1": { |
| "text": "Trend graph for the Major Leagues data set display a graph where regular temporal intervals are used in the temporal axis.", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "FIGREF2": { |
| "text": "Sentence extraction and highlighting display for the typhoon data set", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "FIGREF3": { |
| "text": "0.4737 0.1422 0.1154 0.5746 0.4211 0.7368 0.2127 0.1674 Method 2 0.3877 0.3158 0.4737 0.1422 0.1196 0.5544 0.4211 0.7368 0.2127 0.1723 Method 3 0.3895 0.3158 0.5263 0.1422 0.1202 0.5491 0.4211 0.7895 0.2127 0.1705 Method 4 0.2216 0.1053 0.3684 0.0846 0.0738 0.3765 0.2105 0.5789 0.1328 0", |
| "type_str": "figure", |
| "uris": null, |
| "num": null |
| }, |
| "TABREF0": { |
| "text": "Examples of extracting important expressions", |
| "type_str": "table", |
| "content": "<table><tr><td/><td>Typhoon</td><td/></tr><tr><td>item units</td><td>temporal units</td><td>item expressions</td></tr><tr><td>gou</td><td>nichi</td><td>taihuu</td></tr><tr><td>(No.)</td><td>(day)</td><td>(typhoon)</td></tr><tr><td>me-toru</td><td>ji</td><td>gogo</td></tr><tr><td>(meter(s))</td><td>(o'clock)</td><td>(afternoon)</td></tr><tr><td>nin</td><td>jigoro</td><td>higai</td></tr><tr><td>(people)</td><td>(around x o'clock)</td><td>(damage)</td></tr><tr><td>kiro</td><td>fun</td><td>shashin setsumei</td></tr><tr><td>(kilometer(s))</td><td>(minute(s))</td><td>(photo caption)</td></tr><tr><td>miri</td><td>jisoku</td><td>chuushin</td></tr><tr><td>(millimeter(s))</td><td>(per hour)</td><td>(center)</td></tr><tr><td/><td>Major League</td><td/></tr><tr><td>item units</td><td>temporal units</td><td>item expressions</td></tr><tr><td>gou</td><td>nichi</td><td>Maguwaia</td></tr><tr><td>(No.)</td><td>(day)</td><td>(McGwire)</td></tr><tr><td>hon</td><td>nen</td><td>honruida</td></tr><tr><td>(home run(s))</td><td>(year)</td><td>(home run)</td></tr><tr><td>kai</td><td>gatsu</td><td>Ka-jinarusu</td></tr><tr><td>(inning(s))</td><td>(month)</td><td>(Cardinals)</td></tr><tr><td>honruida</td><td>nen buri</td><td>Ma-ku Maguwaia ichiruishu</td></tr><tr><td>(home run(s))</td><td colspan=\"2\">(after x year(s) interval) (Mark McGwire, the first baseman)</td></tr><tr><td>shiai</td><td>fun</td><td>So-sa</td></tr><tr><td>(game(s))</td><td>(minute(s))</td><td>(Sosa)</td></tr><tr><td/><td>Political Trend</td><td/></tr><tr><td>item units</td><td>temporal units</td><td>item expressions</td></tr><tr><td>%</td><td>gatsu</td><td>naikaku shiji ritsu</td></tr><tr><td>(%)</td><td>(month)</td><td>(cabinet approval rating)</td></tr><tr><td>pointo gen</td><td>nichi</td><td>Obuchi naikaku</td></tr><tr><td colspan=\"2\">(decrease of x point(s)) (day)</td><td>(Obuchi Cabinet)</td></tr><tr><td>pointo zou</td><td>nen</td><td>Obuchi shushou</td></tr><tr><td colspan=\"2\">(increase of x point(s)) (year)</td><td>(Prime Minister Obuchi)</td></tr><tr><td>dai</td><td>kagetu</td><td>shijiritsu</td></tr><tr><td>(generation)</td><td>(month(s))</td><td>(approval rating)</td></tr><tr><td>pointo</td><td>bun no</td><td>kitai</td></tr><tr><td>(point(s))</td><td>(divided)</td><td>(expectation)</td></tr></table>", |
| "num": null, |
| "html": null |
| }, |
| "TABREF2": { |
| "text": "Experimental results for the open data", |
| "type_str": "table", |
| "content": "<table><tr><td/><td colspan=\"2\">Evaluation A</td><td/><td/><td/><td colspan=\"2\">Evaluation B</td><td/><td/></tr><tr><td>MRR</td><td>TP1</td><td>TP5</td><td>RP</td><td>AP</td><td>MRR</td><td>TP1</td><td>TP5</td><td>RP</td><td>AP</td></tr><tr><td/><td/><td colspan=\"5\">Use of Equation 1 and the expression length</td><td/><td/><td/></tr></table>", |
| "num": null, |
| "html": null |
| } |
| } |
| } |
| } |
