Remembrance agent : from theory to applications

• Main Author: Han, Xiaogang
• Other Authors: Miao Chun Yan
• Format: Theses and Dissertations
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Information systems::Information storage and retrieval
• Online Access: https://hdl.handle.net/10356/59381
• Institution: Nanyang Technological University

In the information age, people have difficulty to process and organize large amount of incoming information from online social network sites due to our limited memory capacity. The problem arise when such information is needed as reference to make decisions. People may not be able to figure out proper clues to retrieve them or may even forget their existence at all. Remembrance agents (RAs) are a type of agents which proactively retrieve and present the user's historical documents based on the his/her current context. Memory modeling is one of the fundamental components for RAs. Most of the existing RAs adopt memories with restricted semantic interpretation, which limit content relevance as the primary criterion for retrieval. Building user interest models based on the user documents stored in RA's memory is very critical to inferring user preferences. Through user interest modeling, potentially interesting documents can be recommended to the user whenever needed. To date, little work has been done on integrating user interest model into RAs. A user's documents are naturally distributed in multiple information sources. Each of the sources is of different characteristics and significance. This contradicts the assumptions made by most of the existing work that the user documents are collected from a single information source. To address these challenges, this thesis proposes a novel RA model, Episodic and Semantic memory based Remembrance Agent (ESRA), inspired by cognitive memory theories. The ESRA incorporates the functionalities of episodic memory (EM) and semantic memory (SM) of the human memory into the agent's memory structure. EM stores documents pertaining to the events experienced by the user in chronological order. SM stores the concepts which are extracted from the documents in EM and their corresponding aggregated activation scores. By explicitly modeling EM and SM and their functional interactions, we can capture both the episodic information and categorical semantic information to improve the memory retrieval performance. Moreover, with an ontological user interest profile constructed from the aggregated categorical concepts in SM, the ESRA is able to provide accurate personalized recommendations for the user. In addition, we propose an approach for unifying a user's historical documents from multiple information sources, and develop a source-aware retrieval algorithm subsequently to complement the user's cognition ability with the proposed ESRA in various scenarios. To study how the proposed ESRA can help the user in real world applications, we have incorporated the ESRA into three different application domains. The first application is the modeling of user interests on folksonomy services and its application for personalized search. The second area of application is tweets recommendation based on ontological user interest profiling on microblogging services. And the third area is to provide assistance to students in remembering things in virtual learning environments. Extensive experiments have shown that the proposed ESRA can enhance the user's cognitive ability by acting as an external memory and significantly improve the historical documents retrieval performance for the user.