隣接度に基づく周辺検索〜意外な情報に辿り着く〜
Neighborhood-based Search: Incorporating Page Adjacency to Reach Serendipitous Information

概要

With the help of state-of-the-art information retrieval technologies, most relevant web pages come to the very top of ranked list in the response of user’s query. They are well performed for navigational search because the gap between user’s search intent and his formed query is small. However, for informational search, or more general case, exploratory search, whose search process is iterative, multi-tactical, spanning multiple queries or search sessions, they don’t seem to be effective enough because the gap is too big to guide users towards correct answers. Another vital reason is that “relevance” is considered to be the most crucial feature rather than “relatedness” which contains more, such as rival information, extended topics. In this work, we state a method to present the whole view of related information according to a user-indicated web page in order to show a specific information space through the introduction of “page adjacency”. Generally speaking, “page adjacency” is referred to as the degree that two documents or web pages share the same, or related topics. Here we are aiming at finding not so similar but adjacent pages (to the user-indicated one) rather than most similar ones. Besides, it can also be used in search result diversification as it demands related (generalized relevance) and novel (partially semantic overlapped) documents towards a document query. Finally, a case study based on a news dataset is conducted to demonstrate the usefulness of our proposed method in finding related (or even serendipitous) information. Besides, we also employ the proposed method on Wikipedia to clean up the mass and meaningless links in order to restore the whole view without superfluous information.

産業界への展開例・適用分野

Suppose a company want to promote its new product, i.e. a single-lens reflex camera. From the user’s perspective, he has plenty of single-lens reflex cameras to choose. He will choose the new product only if it outperformed other choices, i.e. high cost performance, or thinner and lighter design. While advertising only states good features and seldom mentions bad ones, that makes information asymmetry. Thus, our potential buyer is likely to spend much time searching to make a survey in order to confirm it does have much more good features. If the buyer could have a glance at "single-lens reflex camera" space in advance, especially provided by the company, not only the promotion will success but also the customer relationship will be improved since customers get both convenient service and objective information.

研究者

氏名	専攻	研究室	役職/学年
趙 夢	社会情報学専攻	田中研究室	博士1回生
大島 裕明	社会情報学専攻	田中研究室	特定准教授
田中 克己	社会情報学専攻	田中研究室	教授

---