Exploring the impact of comments on the blog posts search

An important aspect which distinguishes blogs from other web pages is the ability of blog readers to respond to blog posts directly by posting comments. In this study, we focus on the comments information in the blogs. In particular, we explore the use of comments in two blog posts search tasks. One task is topical relevance retrieval which aims to retrieve as many on-topic posts as possible; the other is opinion retrieval which targets not only topical relevant posts but also the ones that contain an opinion about the given topic. We mainly investigate two aspects of comments: volume and content, focusing on the volume aspect. For both search tasks, we view the comments volume as a feature for ranking documents before we know the query (such a feature is called a query independent feature) and apply two established techniques to incorporate a query independent feature in document ranking. Furthermore, we explore the effect of noisy comments which are either irrelevant to their corresponding post in terms of content or are spam comments on the retrieval performance for both search tasks. As for the comments content, we investigate its impact for both search tasks from two angles: how the retrieval performance will be changed if we omit comments content during retrieval and what is the contribution of comments content in retrieving relevant (or relevant opinionated) documents. As a necessary step for all the above work, we extract the comments from the TREC Blogs06 test collection and study various comments statistics. In addition, we explore a proximity-based technique for opinion retrieval which measures the degree of opinion in a document by the proximity between the query terms and the subjective terms in the document. The experimental results show that using the comments volume feature is generally helpful for both search tasks. However, removing the noisy comments is not helpful for both search tasks, even when varying the level of the noise to be considered. Furthermore, the comments content has an impact on the retrieval performance for both search tasks, verified from both angles. Finally, the proximity-based opinion detection technique only works if an appropriate setting is applied.

Level dc:type.qualificationlevel

MSc(R)

Grantor dc:publisher.institution

University of Glasgow

Year dc:date.issued

2009

Author dc:creator

• Qiu, Sukai

• QA75 Electronic computers. Computer science

DOI dc:identifier.doi

https://doi.org/10.5525/gla.thesis.708

OAI identifier oai:identifier

oai:theses.gla.ac.uk:708