[3Xin4] Poster session 1

[3Xin4-77] Analysis of Search Queries Characteristic of Feline Diseases Using PMI and BERT

〇Naomi Sasaya¹, Tomohide Shibata¹, Shigefumi Kishida², Ryo Kikuchi², Akira Tajima¹ (1.Yahoo Japan Corporation, 2.Anicom Insurance, Inc.)

Keywords:BERT, search query, pet diseases, data science, health care

When a pet becomes ill, owners often wonder whether they should take their pet to a veterinary clinic. Before visiting a veterinary clinic, pet owners search the Internet for their pet’s symptoms and behaviors, and once a diagnosis is made, they search for medications, treatments, and test values. If such owners can be provided with a candidate name of disease at an early stage, it can lead to early detection. Focusing on these search behaviors, this study detects search queries that are characteristic of feline diseases based on Pointwise Mutual Information (PMI) using a large search query log. Since characteristic search queries include not only symptoms and behaviors, but also medications, treatments, test values, and other queries, we classify them into categories using BERT to structure and visualize the search queries.