Joint analysis of user-generated content and product information to enhance user experience in e-commerce

The development of Internet has brought us a more convenient way to purchase goods through e-commerce, which has gradually pervaded our life. However, shopping experience of users in e-commerce has been far from the optimum. In order to enhance user experience in e-commerce, we propose a series of novel studies based on joint analysis of user-generated content and product information; in this dissertation, user-generated content includes user reviews and social media text data, and product information includes product descriptions and product specifications in general. This dissertation aims at assisting e-commerce users in two directions: discovering products and making purchase decisions. To help users discover products, we first propose to leverage user reviews to improve accuracy of product search. We carefully combine product descriptions and user reviews to improve product search. Then, we also propose to recommend products via inference of implicit intent in social media text. We infer implicit intent in user status text leveraging parallel corpora we build from social media, and we recommend products whose descriptions satisfy the inferred intent. In order to help users make purchase decisions, we first propose to generate augmented product specifications leveraging user reviews. Product specifications are often difficult to understand especially for high-technology products that contain many advanced features. We jointly model user reviews and product specifications to augment product specifications with useful information in the user reviews. We also propose to retrieve relevant opinions for new products. New or unpopular products often have no reviews, and such lack of information makes consumers hesitate to make a purchase decision. We leverage user reviews of similar products, where similarity is estimated using product specifications, to retrieve relevant opinions for new products. The experiment results show the proposed models are effective in general. The models are also general enough to be applied to any entities with their text data. Furthermore, the models can benefit both product manufacturers and consumers, so their potential impact may be even bigger.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Park, Dae Hoon

Contributors dc:contributor

• Zhai, ChengXiang
• Han, Jiawei
• Chang, Kevin C.C.
• Fang, Yi

• Information Retrieval
• Text Mining
• E-commerce
• User Review Mining

Statement dc:rights

• Copyright 2016 Dae Hoon Park

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/90602

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/90602