Trend Dataset: Evaluation Datasets for Trend Forecasting Studies

<p><strong><a href="https://ojs.aaai.org/index.php/ICWSM/article/view/22212">A conference&nbsp;paper</a>&nbsp;has published in AAAI-ICWSM 2023&nbsp;details the method to create the data.</strong></p> <p><strong>Overview</strong></p> <p>This dataset is created for the purpose of evaluating trend forecasting methods. The dataset contains 400 entities in 21 domains consisting of 5 categories (geography, organization, person, product, and content). Each entity is annotated with three types of trend attributes: trending status (trending or non-trending), degree of trending (how well it is recognized), and trend period (when the trend started and ended).</p> <p>There are three main features of this dataset. First, a questionnaire-based recognition rate is used as the gold standard for annotating trend attributes. Second, a collection of entities from a wide range of domains while covering both trending and non-trending, without significant imbalance. Third, trend period annotation on a weekly resolution through interpolation using Internet search volume data.</p> <p>See our paper &quot;Construction of Evaluation Datasets for Trend Forecasting Studies&quot; *1 (hereafter known as &ldquo;paper&rdquo;) for details.</p> <p>*1 This paper is currently under review in single blind.</p> <p><strong>Conditions under this dataset construction</strong></p> <p><strong>Target Trending Phenomenon</strong></p> <table> <thead> <tr> <th>attribute</th> <th>value</th> </tr> </thead> <tbody> <tr> <td>Target Country</td> <td>Japan</td> </tr> <tr> <td>Survey Period</td> <td>from 2015 to 2019</td> </tr> <tr> <td># of entities</td> <td>400</td> </tr> <tr> <td>Target Domains</td> <td>21 domains of 5 categories (See Target Domain Section)</td> </tr> </tbody> </table> <p><strong>Target Domain</strong></p> <table> <thead> <tr> <th>Category</th> <th>Domain</th> </tr> </thead> <tbody> <tr> <td>Location/Geography</td> <td>City/region/landmark</td> </tr> <tr> <td>Organization</td> <td>Restaurant/facility, company/brand</td> </tr> <tr> <td>Person/Group</td> <td>Politician/political party, researcher, athlete, actor/actress, celebrity/entertainer/comedian, music band/music group</td> </tr> <tr> <td>Products</td> <td>Cosmetics, daily necessities, clothing, beverage, foodstuff, others</td> </tr> <tr> <td>Art/Content</td> <td>Game, publication, comic/animation, movie, broadcast program, music</td> </tr> </tbody> </table> <p><strong>Dataset Files</strong></p> <p>The dataset contains two TSV files and a single Markdown file. The description and metadata about the dataset are provided in a Markdown file. The list of files is shown in the table below.</p> <table> <thead> <tr> <th>data label</th> <th>file name</th> <th>file format</th> <th>explanation</th> </tr> </thead> <tbody> <tr> <td>Metadata</td> <td>README.md</td> <td>Markdown</td> <td>Overview of this dataset and the schema of each file in English and Japanese</td> </tr> <tr> <td>Trend Dataset</td> <td>trend_dataset.tsv</td> <td>TSV</td> <td>Main body of this dataset</td> </tr> <tr> <td>Master Entity list</td> <td>master_entity_list.tsv</td> <td>TSV</td> <td>Information of entities included in Trend Dataset</td> </tr> </tbody> </table> <p><strong>TSV File format</strong></p> <table> <thead> <tr> <th>attribute</th> <th>value</th> </tr> </thead> <tbody> <tr> <td>header row</td> <td>exists (first row)</td> </tr> <tr> <td>index column</td> <td>&ldquo;ID&rdquo; column (first column)</td> </tr> <tr> <td>encoding</td> <td>UTF-8</td> </tr> <tr> <td>delimiter</td> <td>\t (tab)</td> </tr> <tr> <td>quoting</td> <td>None *2</td> </tr> <tr> <td>escape character</td> <td>\ (back slash)</td> </tr> <tr> <td>line terminator</td> <td>\n</td> </tr> </tbody> </table> <p>*2: <a href="https://docs.python.org/3/library/csv.html#csv.QUOTE_NONE">csv.QUOTE_NONE</a> in Python</p>