Kazunari Sugiyama

Adaptive web search based on user profile constructed without any effort from users

Web search engines help users find useful information on the World Wide Web (WWW). However, when the same query is submitted by different users, typical search engines return the same result regardless of who submitted the query. Generally, each user has different information needs for his/her query. Therefore, the search result should be adapted to users with different information needs. In this paper, we first propose several approaches to adapting search results according to each users need for relevant information without any user effort, and then verify the effectiveness of our proposed approaches. Experimental results show that search systems that adapt to each users preferences can be achieved by constructing user profiles based on modified collaborative filtering with detailed analysis of users browsing history in one day.

Addressing cold-start in app recommendation: latent user models constructed from twitter followers

As a tremendous number of mobile applications (apps) are readily available, users have difficulty in identifying apps that are relevant to their interests. Recommender systems that depend on previous user ratings (i.e., collaborative filtering, or CF) can address this problem for apps that have sufficient ratings from past users. But for apps that are newly released, CF does not have any user ratings to base recommendations on, which leads to the cold-start problem. In this paper, we describe a method that accounts for nascent information culled from Twitter to provide relevant recommendation in such cold-start situations. We use Twitter handles to access an apps Twitter account and extract the IDs of their Twitter-followers. We create pseudo-documents that contain the IDs of Twitter users interested in an app and then apply latent Dirichlet allocation to generate latent groups. At test time, a target user seeking recommendations is mapped to these latent groups. By using the transitive relationship of latent groups to apps, we estimate the probability of the user liking the app. We show that by incorporating information from Twitter, our approach overcomes the difficulty of cold-start app recommendation and significantly outperforms other state-of-the-art recommendation techniques by up to 33%.

Refinement of TF-IDF schemes for web pages using their hyperlinked neighboring pages

In IR (information retrieval) systems based on the vector space model, the TF-IDF scheme is widely used to characterize documents. However, in the case of documents with hyperlink structures such as Web pages, it is necessary to develop a technique for representing the contents of Web pages more accurately by exploiting the contents of their hyperlinked neighboring pages. In this paper, we first propose several approaches to refining the TF-IDF scheme for a target Web page by using the contents of its hyperlinked neighboring pages, and then compare the retrieval accuracy of our proposed approaches. Experimental results show that, generally, more accurate feature vectors of a target Web page can be generated in the case of utilizing the contents of its hyperlinked neighboring pages at levels up to second in the backward direction from the target page.

New and improved: modeling versions to improve app recommendation

Existing recommender systems usually model items as static -- unchanging in attributes, description, and features. However, in domains such as mobile apps, a version update may provide substantial changes to an app as updates, reflected by an increment in its version number, may attract a consumers interest for a previously unappealing version. Version descriptions constitute an important recommendation evidence source as well as a basis for understanding the rationale for a recommendation. We present a novel framework that incorporates features distilled from version descriptions into app recommendation. We use a semi-supervised topic model to construct a representation of an apps version as a set of latent topics from version metadata and textual descriptions. We then discriminate the topics based on genre information and weight them on a per-user basis to generate a version-sensitive ranked list of apps for a target user. Incorporating our version features with state-of-the-art individual and hybrid recommendation techniques significantly improves recommendation quality. An important advantage of our method is that it targets particular versions of apps, allowing previously disfavored apps to be recommended when user-relevant features are added.

Product review summarization from a deeper perspective

With product reviews growing in depth and becoming more numerous, it is growing challenge to acquire a comprehensive understanding of their contents, for both customers and product manufacturers. We built a system that automatically summarizes a large collection of product reviews to generate a concise summary. Importantly, our system not only extracts the review sentiments but also the underlying justification for their opinion. We solve this problem through a novel application of clustering and validate our approach through an empirical study, obtaining good performance as judged by F-measure (the harmonic mean of purity and inverse purity).

Serendipitous recommendation for scholarly papers considering relations among researchers

Serendipity occurs when one finds an interesting discovery while searching for something else. While search engines seek to report work relevant to a targeted query, recommendation engines are particularly well-suited for serendipitous recommendations as such processes do not need to fulfill a targeted query. Junior researchers can use such an engine to broaden their horizon and learn new areas, while senior researchers can discover interdisciplinary frontiers to apply integrative research. We adapt a state-of-the-art scholarly paper recommendation systems user profile construction to make use of information drawn from 1) dissimilar users and 2) co-authors to specifically target serendipitous recommendation.

Serendipitous Recommendation for Mobile Apps Using Item-Item Similarity Graph

Recommender systems can provide users with relevant items based on each user’s preferences. However, in the domain of mobile applications (apps), existing recommender systems merely recommend apps that users have experienced (rated, commented, or downloaded) since this type of information indicates each user’s preference for the apps. Unfortunately, this prunes the apps which are releavnt but are not featured in the recommendation lists since users have never experienced them. Motivated by this phenomenon, our work proposes a method for recommending serendipitous apps using graph-based techniques. Our approach can recommend apps even if users do not specify their preferences. In addition, our approach can discover apps that are highly diverse. Experimental results show that our approach can recommend highly novel apps and reduce over-personalization in a recommendation list.

A comprehensive evaluation of scholarly paper recommendation using potential citation papers

To help generate relevant suggestions for researchers, recommendation systems have started to leverage the latent interests in the publication profiles of the researchers themselves. While using such a publication citation network has been shown to enhance performance, the network is often sparse, making recommendation difficult. To alleviate this sparsity, in our former work, we identified “potential citation papers” through the use of collaborative filtering. Also, as different logical sections of a paper have different significance, as a secondary contribution, we investigated which sections of papers can be leveraged to represent papers effectively. While this initial approach works well for researchers vested in a single discipline, it generates poor predictions for scientists who work on several different topics in the discipline (hereafter, “intra-disciplinary”). We thus extend our previous work in this paper by proposing an adaptive neighbor selection method to overcome this problem in our imputation-based collaborative filtering framework. On a publicly-available scholarly paper recommendation dataset, we show that recommendation accuracy significantly outperforms state-of-the-art recommendation baselines as measured by nDCG and MRR, when using our adaptive neighbor selection method. While recommendation performance is enhanced for all researchers, improvements are more marked for intra-disciplinary researchers, showing that our method does address the targeted audience.

Identifying citing sentences in research papers using supervised learning

Researchers have largely focused on analyzing citation links from one scholarly work to another. Such citing sentences are an important part of the narrative in a research article. If we can automatically identify such sentences, we can devise an editor that helps suggest when a particular piece of text needs to be backed up with a citation or not. In this paper, we propose a method for identifying citing sentences by constructing a classifier using supervised learning. Our experiments show that simple language features such as proper nouns and the labels of previous and next sentences are effective features to identifying citing sentences.

Instructor-centric source code plagiarism detection and plagiarism corpus

Existing source code plagiarism systems focus on the problem of identifying plagiarism between pairs of submissions. The task of detection, while essential, is only a small part of managing plagiarism in an instructional setting. Holistic plagiarism detection and management requires coordination and sharing of assignment similarity -- elevating plagiarism detection from pairwise similarity to cluster-based similarity; from a single assignment to a sequence of assignments in the same course, and even among instructors of different courses. To address these shortcomings, we have developed Student Submissions Integrity Diagnosis (SSID), an open-source system that provides holistic plagiarism detection in an instructor-centric way. SSIDs visuals show overviews of plagiarism clusters throughout all assignments in a course as well as highlighting most-similar submissions on any specific student. SSID supports plagiarism detection workflows; e.g., allowing student assistants to flag suspicious assignments for later review and confirmation by an instructor with proper authority. Evidence is automatically entered into SSIDs logs and shared among instructors. We have additionally collected a source code plagiarism corpus, which we employ to identify and correct shortcomings of previous plagiarism detection engines and to optimize parameter tuning for SSID deployment. Since its deployment, SSIDs workflow enhancements have made plagiarism detection in our faculty less tedious and more successful.