Update README.md

README.md

@@ -21,7 +21,7 @@ We envision a scenario in which the human and the robot engage in discussion ove
 
 From the above illustration, we can deduce that the phrase `Cochinita Pibil for sure!` is related to the `intent` called `food` because the human is answering the question `what's your favorite cuisine?` 
 
-## :card_file_box: The Dataset
+## 🗂️ 🗂️ The Dataset
 In this dataset, we set out to defer from the conventional norm in which intent classification datasets are constructed.
 For each sentence in the dataset, only a label identifying the intent label to which the sentence belongs is included. 
 
@@ -43,7 +43,7 @@ The file named *HRI_TOTAL_data.csv* contains all of the data found in the 6 file
 On top of that, we have provided more specific data files corresponding to `four languages`; English (En), Japanese (Jp), Swahili (Sw), Urdu (Ur), and `six intent classes` such as `IntentRecognitionData_En_Intent_Sports.csv`
 
 Feel free to use whichever data files you are interested in, from the `./data/` folder.
-## :dizzy: The Data format
+## 💫 💫 The Data format
 We have provided the data in tabular format with two columns. Column 1 contains the **Sentence** while column 2 contains the **Intent_label**. 
 
 <table>
@@ -77,7 +77,7 @@ We have provided the data in tabular format with two columns. Column 1 contains
   </tr>
 </table>
 
-## :mechanical_arm: The Prompts are shown below.
+## 🦾🦾 The Prompts are shown below.
 Here are the prompts used in the experiments.
 
 #### 1. Zero-shot Standard
@@ -229,7 +229,7 @@ Tell me how many predictions you make in total.d
 ```
 
 The prompts above facilitated our **zero-shot** intent classification analysis.
-## :sparkles: Evaluation
+## ✨✨ Evaluation
 We conducted experiments with data sizes per intent class of `200`, `500`, and `all data`. Moreover, we used the three models: `Gemma`, `Claude-3-Opus`, and `GPT-4-turbo`.
 <!--
 🚧 **Currently under construction....** 🚧