Takehiro Yamamoto

本研究では，提示型検索モデル（Ostensive Search Model）に基づくインタフェースによって鑑賞者個人の興味を反映したミュージアム体験を可能にする電子ガイドを提案し，そのログを分析することで実現可能になった事前学習，事後学習支援システムについても提案する．我々は，国立民族学博物館（みんぱく）の展示物のうち3,053点について，展示物の解説やビデオなどの詳細情報を検索し，メモなどのアノテーションを付与できるiPad用アプリケーションである「みんぱくガイド」を作成した．みんぱくガイドは，鑑賞者が新しい展示物に気付いたり，興味を明確化することができるように，一覧性の高い検索結果画面を中心に情報探索を繰り返せるインタフェースを持っている．このような電子ガイドの操作履歴や位置情報などのログを用いることで，個人のミュージアム体験を色濃く反映した事前・事後学習支援を可能にした．事前学習支援システムでは，ミュージアムに行く前にカードを整理しながら鑑賞計画を立てるウェブアプリケーションの利用を通じて，鑑賞者に自分が何を学びに行くかという鑑賞軸を自覚してもらう．また，事後学習を促す仕組みとして，ログから鑑賞者が興味を持った展示物を推定し，後から鑑賞体験を思い出しやすくするためのパーソナライズされたポストカードを自動生成するシステムも作成した． This paper proposes an electronic guide application that enables visitors to get a museum experience that reflects their individual interests through an interface based on an Ostensive Search Model. In addition, we propose a pre-learning and post-learning support system that can be connected to our electronic guide. The system is based on a search result screen with a high level of browsability. We propose a system that allows users to search for detailed information such as explanatory texts and videos of exhibits they are interested in and add annotations such as handwritten notes. We created the “Minpaku Guide,” an iPad application that allows users to search for detailed information such as explanatory text and videos on exhibits of interest and add annotations such as notes. We also developed pre-learning support systems connected to our guide application. As a pre-learning support system, we created a web application that allows users to manually organize cards before visiting the museum to clarify what the visitors will learn beforehand. As a post-learning support system, we implemented a system that summarizes the visitor's operation log of the guide application into a postcard that helps to recall the viewing experience later.

Web searchers often use a Web search engine to find a way or means to achieve his/her goal. For example, a user intending to solve his/her sleeping problem, the query “sleeping pills” may be used. However, there may be another solution to achieve the same goal, such as “have a cup of hot milk” or “stroll before bedtime.” The problem is that the user may not be aware that these solutions exist. Thus, he/she will probably choose to take a sleeping pill without considering these solutions. In this study, we define and tackle the alternative action mining problem. In particular, we attempt to develop a method for mining alternative actions for a given query. We define alternative actions as actions which share the same goal and define the alternative action mining problem as similar in the search result diversification. To tackle the problem, we propose leveraging a community Q&amp A (cQA) corpus for mining alternative actions. The cQA corpus can be seen as an archival dataset comprising dialogues between questioners, who want to know the solutions to their problem, and respondents, who suggest different solutions. We propose a method to compute how well two actions can be alternative actions by using a question-answer structure in a cQA corpus. Our method builds a question-action bipartite graph and recursively computes how well two actions can be alternative actions. We conducted experiments to investigate the effectiveness of our method using two newly built test collections, each containing 50 queries. The experimental results indicated that, for Japanese test collection, our proposed method significantly outperformed two types of baselines, one used the conventional query suggestions and the other extracted alternative-actions from the Web documents, in terms of D#-nDCG@8. Also, for English test collection, our method significantly outperformed the baseline using the conventional query suggestions in terms of D#-nDCG@8.

Web searchers often use a Web search engine to find a way or means to achieve his/her goal. For example, a user intending to solve his/her sleeping problem, the query "sleeping pills" may be used. However, there may be another solution to achieve the same goal, such as "have a cup of hot milk" or "stroll before bedtime." The problem is that the user may not be aware that these solutions exist. Thus, he/she will probably choose to take a sleeping pill without considering these solutions. In this study, we define and tackle the alternative action mining problem. In particular, we attempt to develop a method for mining alternative actions for a given query. We define alternative actions as actions which share the same goal and define the alternative action mining problem as similar in the search result diversification. To tackle the problem, we propose leveraging a community Q&amp A (cQA) corpus for mining alternative actions. We propose a method to compute how well two actions can be alternative actions by using a question-Answer structure in a cQA corpus. Our method builds a question-Action bipartite graph and recursively computes how well two actions can be alternative actions.We conducted experiments to investigate the effectiveness of our method using two newly built test collections, each containing 50 queries. The experimental results indicated that our proposed method outperformed the query suggestion methods provided by the commercial search engines in terms of D-nDCG.

This study proposes a method for retrieving and ranking posts from social network services(SNSs) by specifying and providing feedback on the context of posts. Current search systems for SNS posts cannot handle user intent with regard to the context of posts to be retrieved, mainly owing to the incompleteness of SNS posts, i.e., they do not contain the users' contexts (e.g., situations or preferences) of users posting messages. Hence, we propose a search method that accepts two kinds of queries, namely, content queries and context queries, and that updates these queries based on the user feedback with special attention to the contexts of posts. Our search method considers the whole SNS dataset as a graph and the nodes surrounding each post as its context to find relevant posts in terms of content and context, our method propagates user feedback via this graph. Our experimental results based on a Twitter test collection revealed that our proposed method showed improved retrieval performance as compared with conventional SNS retrieval and relevance feedback. In addition, we could detect the optimal parameters for feedback propagating.

We present one of the world's first a.empts to examine the feasibility of multileaving evaluation of document rankings on a large scale commercial community .Thestion Answering (cQA) service. As a natural enhancement of interleaving evaluation, multileaving merges more than two input rankings into one and measures the search user satisfaction of each input ranking on the basis of user clicks on the multileaved ranking. We evaluated the adequateness of two major multileaving methods, team dra. multileaving (TDM) and optimized multileaving (OM), proposing their practical implementation for live services. Our experimental results demonstrated that multileaving methods could precisely evaluate the effectiveness of five rankings with di.erent quality by using clicks from real users. Moreover, we concluded that OM is more efficient than TDM by observing that most of the evaluation results with OM converged a.er showing multileaved rankings around 40,000 times and an in-depth analysis of their characteristics.

We address the problem of event identification on microblogs with special attention to implicit reference cases in which events are not referred to by event’s information. Most studies identify events referred to by event’s information, while there are many implicitly referred events by microblogs, which are difficult to identify for short text such as microblogs. We therefore tackled implicit reference cases by analyzing links from microblogs. The links are able to connect opinions or feeling to their referred events. The analysis of links is particularly important for certain types of implicit references. In addition, we predict reference type of a microblog for accurately ranking referred events. The experimental results suggest that our method was effective for implicit references and predicting reference type was essential for identifying implicitly or explicitly referred events together.

In recall-oriented tasks, where collecting extensive information from different aspects of a topic is required, searchers often have difficulty formulating queries to explore diverse aspects and deciding when to stop searching. With the goal of helping searchers discover unexplored aspects and find the appropriate timing for search stopping in recall-oriented tasks, this paper proposes a query suggestion interface displaying the amount of missed information (i.e., information that a user potentially misses collecting from search results) for individual queries. We define the amount of missed information for a query as the additional gain that can be obtained from unclicked search results of the query, where gain is formalized as a set-wise metric based on aspect importance, aspect novelty, and per-aspect document relevance and is estimated by using a state-of-the-art algorithm for subtopic mining and search result diversification. Results of a user study involving 24 participants showed that the proposed interface had the following advantages when the gain estimation algorithm worked reasonably: (1) users of our interface stopped examining search results after collecting a greater amount of relevant information (2) they issued queries whose search results contained more missed information (3) they obtained higher gain, particularly at the late stage of their sessions and (4) they obtained higher gain per unit time. These results suggest that the simple query visualization helps make the search process of recall-oriented tasks more efficient, unless inaccurate estimates of missed information are displayed to searchers.

In this paper, we propose a method to find query suggestions of a verbal query, which contains a verb in the query, from the Web. People sometimes cannot obtain appropriate search results even if they consider they have formulated a query that clearly describes their search intents. The idea of the proposed method is to find the relationship between verb and noun in the query, and mine the appropriate representation of the verb based on the relationship. The proposed method estimates the relationship between verb and noun based on particles between them. Based on the estimated relationship, we then obtain candidates of the verb in the query by using either the Web search results or the case frame. Next, we compute the effectiveness of the candidates by considering the similarity between a candidate and the verb and the co-occurrence between the candidates and the noun, and finally rank the candidates to generate queries. To investigate the effectiveness of our proposed method, we conducted the experiment by comparing with the query suggestions of a commercial search engine as our baseline. The experimental result of 20 queries showed that our proposed method, which finds candidates from the Web search results, outperformed the baseline method in terms of AvgRelNum, which measures the the number of relevant pages obtained by the generated query that can retrieve a relevant page, and achieved the similar performance in terms of Contain@10 and MRR@10.

In this paper, we propose a method of ranking entities (e.g. products) based on pairwise preferences learned and inferred from user reviews. Our proposed method finds expressions from user reviews that indicate pairwise preferences of entities in terms of a certain attribute, and learns a function that determines the relative degree of the attribute to rank entities. Since there are a limited number of such expressions in reviews, we further propose a method of inferring pairwise preferences based on attribute dependencies obtained from reviews. As some pairwise preferences are less confident, we also propose a modified version of a learning to rank method, Fuzzy Ranking SVM, which can take into account the uncertainty of pairwise preferences. The experiment was carried out with three categories of products and several attributes specific to each category. The experimental results showed that our approach could learn more accurate pairwise preferences than baseline methods, and inference based on the attribute dependency could improve the performances.

We address the problem of entity identification on a microblog with special attention to indirect reference cases in which entities are not referred to by their names. Most studies on identifying entities referred to them by their full/partial name or abbreviation, while there are many indirectly mentioned entities in microblogs, which are difficult to identify in short text such as microblogs. We therefore tackled indirect reference cases by developing features that are particularly important for certain types of indirect references and modeling dependency among referred entities by a Conditional Random Field ( CRF) model. In addition, we model non-sequential order dependency while keeping the inference tractable by dynamically building dependency among entities. The experimental results suggest that our features were effective for indirect references, and our CRF model with adaptive dependency was robust even when there were multiple mentions in a microblog and achieved the same high performance as that with the fully connected CRF model.

For intrinsically diverse tasks, in which collecting extensive information from different aspects of a topic is required, searchers often have difficulty formulating queries to explore diverse aspects and deciding when to stop searching. With the goal of helping searchers discover unexplored aspects and find the appropriate timing for search stopping in intrinsically diverse tasks, we propose ScentBar, a query suggestion interface visualizing the amount of important information that a user potentially misses collecting from the search results of individual queries. We define the amount of missed information for a query as the additional gain that can be obtained from unclicked search results of the query, where gain is formalized as a set-wise metric based on aspect importance, aspect novelty, and per-aspect document relevance and is estimated by using a state-of-the-art algorithm for subtopic mining and search result diversification. Results of a user study involving 24 participants showed that the proposed interface had the following advantages when the gain estimation algorithm worked reasonably: (1) ScentBar users stopped examining search results after collecting a greater amount of relevant information (2) they issued queries whose search results contained more missed information (3) they obtained higher gain, particularly at the late stage of their sessions and (4) they obtained higher gain per unit time. These results suggest that the simple query visualization helps make the search process of intrinsically diverse tasks more efficient, unless inaccurate estimates of missed information are visualized.

While search engines sometimes return different documents containing contradictory answers, little is known about how users handle inconsistent information. This paper investigates the effect of search expertise (defined as specialized knowledge on the internal workings of search engines) on search behavior and satisfaction criteria of users. We selected four tasks comprising factoid questions with inconsistent answers, extracted answers that 30 study participants had found in these tasks, and analyzed their answer-finding behavior in terms of the presence or absence of search expertise. Our main findings are as follows: (1) finding inconsistent answers causes users with search expertise (search experts) to feel dissatisfied, while effort in searching for answers is the dominant factor in task satisfaction for those without search expertise (search nonexperts) (2) search experts tend to spend longer completing tasks than search non-experts even after finding possible answers and (3) search experts narrow down the scope of searches to promising answers as time passes as opposed to search non-experts, who search for any answers even in the closing stage of task sessions. These findings suggest that search non-experts tend to be less concerned about the consistency in their found answers, on the basis of which we discuss the design implications for making search non-experts aware of the existence of inconsistent answers and helping them to search for supporting evidence for answers.

We propose a method to generate effective query suggestions aiming to help struggling search, where users experience difficulty in locating information that is relevant to their information need in the search session. The core is identifying struggling component of an on-going struggling session and mining the effective representations of it. The struggling component is the semantic component of information need for which the user struggled to find an effective representation during the struggling session. The proposed method identifies the struggling component of given on-going struggling session and mines the sessions containing the identified struggling component from a query log to build a struggling flow graph. The struggling flow graph records users' reformulation behaviors for the terms of the struggling component, through struggling flow graph we can mine effective representations of the struggling component. The experimental results demonstrate that the proposed method outperforms the baseline methods when it can use two or more queries in a struggling session.