Interest Recognition from Online Instant Messaging Sessions Using Text Segmentation and Document Embedding Techniques
Hana Lee, Young Yoon · 2018
In this paper, we present techniques for recognizing users' interest from online instant messages for more timely and user-friendly targeted advertisement. We devise three text segmentation methods to identify blocks of utterances in an instant messaging session that resonate interests in certain products. We adapt document embedding technique for classifying a given text segment into a multi-level product category. We use over 50,000 product descriptions available on Groupon as training data and evaluate the effectiveness of our approaches based on the chat sessions that are simulated with dialogue from TV drama scripts.