Professor Yoshitaka Kameya provides an explanation.
Based on music reviews, recommending songs that match your preferences.
Recommendation systems are now ubiquitous in music and for books, clothing, real estate properties, and even friends, suggesting personalized options for us in various domains. The mechanism behind these systems is relatively straightforward, relying on machine learning techniques to predict whether a user is likely to purchase an item based on their purchase history, the content of purchased items, and demographic information such as age and gender. Two methods commonly used in these predictions are “item-based collaborative filtering” and “user-based collaborative filtering.” Item-based collaborative filtering identifies and recommends items similar to the ones currently being viewed by the user, and this approach is reportedly employed by some of the world’s leading e-commerce companies. In contrast, user-based collaborative filtering identifies users in the database who have similar purchase histories or attributes to the current user and recommends products those similar users have purchased. While both systems are relatively straightforward, they require vast data processing.
My lab uses these techniques to build a music recommendation system. We leverage CD databases that include song comments provided by music publishers as part of the decision-making process. Today, people can easily enjoy music on platforms like YouTube, but as a student, I would check each POP comment in record stores and imagine what the album might be like before purchasing it. From this experience, I thought it might be possible to recommend songs using recommendation comments and review texts.
In this system, users can input a song they like, and the system will reference reviews from a database containing information on approximately 50,000 songs. It will search for songs that are described using the same words or that have a similar impression. Users can decide and set whether to prioritize lyrics or music first. Based on this setting, the system will recommend songs with similar lyrics content, and mood, or those with a similar melody and tempo. Currently, the system suggests a few songs, but we aim to create playlists that recommend multiple songs as part of a meaningful story, providing a valuable music experience.
Applying Big Data Analysis to the Medical Field
Data analysis can be applied in various fields. For example, some students use statistical methods to analyze sports data under the keyword “data analysis.” In contrast, others employ a method called “cluster analysis,” which finds similar data groups within large datasets to analyze shoppers’ behavior and contribute to revitalizing shopping districts.
Currently, I am researching the application of big data analysis for pattern discovery in the medical field. Recently, polypharmacy, where multiple medications prescribed by different medical institutions and departments lead to adverse effects, has become a concern. This situation can lead to a “prescription cascade,” where side effects are mistaken for new symptoms, resulting in further medication prescriptions. Certain combinations of drugs can cause hypotension, leading to dizziness, falls, and even permanent bedridden states. Therefore, in collaboration with the National Center for Geriatrics and Gerontology, we are analyzing past case data to identify patterns in drug combinations that lead to hypotension. By allowing pharmacists to review prescriptions, we aim to prevent polypharmacy and contribute to the appropriate use of medications.
Interview Date: February 13, 2021