Implicit feedback is a form a relevance feedback that is inferred from how users interact with an information retrieval system such as an online search engine. This degree project report describes a method of using implicit feedback to establish relevance judgments and rank products based on their relevance to a specified attribute. The report contains an overview of the benefits and limitations of implicit feedback, as well as a description on how those limitations can be mitigated. A prototype that interpreted user actions as relevance votes and calculat-ed a fair relevance score based on these votes with the help of an algo-rithm was developed. This system was then tested on a website with real users during a limited period of time. The results from the test period were evaluated and the system was concluded to be far from perfect, but that improvements could be made by making adjustments to the algo-rithm. The system performed better when looking at the algorithm’s pre-cision rather than its sensitivity. / Implicit feedback är en sorts relevansfeedback som sammanställs utifrån användares interaktion med ett informationsökningsssystem. Denna examensarbetesrapport beskriver ett sätt att använda implicit feedback för att skapa en bedömning av en produkts relevans till ett angivet attribut. Rapporten innehåller också en överblick av fördelarna och nackdelarna med implicit feedback, samt en beskrivning av hur dessa nackdelar kan hanteras. En prototyp som översatte användarbeteende till olika relevansröster och beräknade ett relevansvärde baserat på dessa relevansröster med hjälp av en algoritm, utvecklades. Denna prototyp testades sedan på en hemsida med verkliga användare under en begränsad tid. Resultatet från denna testperiod analyserades och gav slutsatsen att prototypen inte var perfekt, men att resultaten kunde förbättras med hjälp av finjusteringar av algoritmen. Prototypens precision, med avseende på vilka produkter algoritmen valde ut som relevanta, var dock bättre än dess sensitivitet.

Doctor of Philosophy / Computing and Information Sciences / Doina Caragea / Increasing amounts of content on the Web means that users can select from a wide variety of items (i.e., items that concur with their tastes and requirements). The generation of personalized item suggestions to users has become a crucial functionality for many web applications as users benefit from being shown only items of potential interest to them. One popular solution to creating personalized item suggestions to users is recommender systems. Recommender systems can address the item recommendation task by utilizing past user preferences for items captured as either explicit or implicit user feedback. Numerous collaborative filtering (CF) approaches have been proposed in the literature to address the recommendation problem in the single-domain setting (user preferences from only one domain are used to recommend items). However, increasingly large datasets often prevent experimentation of every approach in order to choose the one that best fits an application domain. The work in this dissertation on the single-domain setting studies two CF algorithms, Adsorption and Matrix Factorization (MF), considered to be state-of-the-art approaches for implicit feedback and suggests that characteristics of a domain (e.g., close connections versus loose connections among users) or characteristics of data available (e.g., density of the feedback matrix) can be useful in selecting the most suitable CF approach to use for a particular recommendation problem. Furthermore, for Adsorption, a neighborhood-based approach, this work studies several ways to construct user neighborhoods based on similarity functions and on community detection approaches, and suggests that domain and data characteristics can also be useful in selecting the neighborhood approach to use for Adsorption. Finally, motivated by the need to decrease computational costs of recommendation algorithms, this work studies the effectiveness of using short-user histories and suggests that short-user histories can successfully replace long-user histories for recommendation tasks. Although most approaches for recommender systems use user preferences from only one domain, in many applications, user interests span items of various types (e.g., artists and tags). Each recommendation problem (e.g., recommending artists to users or recommending tags to users) can be considered unique domains, and user preferences from several domains can be used to improve accuracy in one domain, an area of research known as cross-domain recommender systems. The work in this dissertation on cross-domain recommender systems investigates several limitations of existing approaches and proposes three novel approaches (two Adsorption-based and one MF-based) to improve recommendation accuracy in one domain by leveraging knowledge from multiple domains with implicit feedback. The first approach performs aggregation of neighborhoods (WAN) from the source and target domains, and the neighborhoods are used with Adsorption to recommend target items. The second approach performs aggregation of target recommendations (WAR) from Adsorption computed using neighborhoods from the source and target domains. The third approach integrates latent user factors from source domains into the target through a regularized latent factor model (CIMF). Experimental results on six target recommendation tasks from two real-world applications suggest that the proposed approaches effectively improve target recommendation accuracy as compared to single-domain CF approaches and successfully utilize varying amounts of user overlap between source and target domains. Furthermore, under the assumption that tuning may not be possible for large recommendation problems, this work proposes an approach to calculate knowledge aggregation weights based on network alignment for WAN and WAR approaches, and results show the usefulness of the proposed solution. The results also suggest that the WAN and WAR approaches effectively address the cold-start user problem in the target domain.

abstract: Recommender systems are a type of information filtering system that suggests items that may be of interest to a user. Most information retrieval systems have an overwhelmingly large number of entries. Most users would experience information overload if they were forced to explore the full set of results. The goal of recommender systems is to overcome this limitation by predicting how users will value certain items and returning the items that should be of the highest interest to the user. Most recommender systems collect explicit user feedback, such as a rating, and attempt to optimize their model to this rating value. However, there is potential for a system to collect implicit user feedback, such as user purchases and clicks, to learn user preferences. Additionally with implicit user feedback, it is possible for the system to remember the context of user feedback in terms of which other items a user was considering when making their decisions. When considering implicit user feedback, only a subset of all evaluation techniques can be used. Currently, sufficient evaluation techniques for evaluating implicit user feedback do not exist. In this thesis, I introduce a new model for recommendation that borrows the idea of opportunity cost from economics. There are two variations of the model, one considering context and one that does not. Additionally, I propose a new evaluation measure that works specifically for the case of implicit user feedback. / Dissertation/Thesis / M.S. Computer Science 2012

The research aims to focus on Recurrent Neural Networks (RNN) and its application to the session-aware recommendations empowered by implicit user feedback and content-based metadata. To investigate the promising architecture of RNN, we implement seven different models utilizing various types of implicit feedback and content information. Our results showed that using RNN with complex implicit feedback increases the next-item prediction comparing the baseline models like Cosine Similarity, Doc2Vec, and Item2Vec.

A growing body of research has compared the effects of explicit and implicit corrective feedback (CF) on L2 accuracy. However, L2 performance is not limited to accuracy. Fluency is another important aspect of L2 performance, but less is understood about its relationship with CF and CF explicitness/implicitness. This experimental study examined the effects of explicit correction versus implicit recasts on not only the accuracy but also the fluency of L2 speech during videoconferencing. Forty-eight lower-intermediate learners of English as a foreign language (EFL) were assigned to an explicit correction group, an implicit recast group, and a no-feedback group. Each engaged in eight picture description tasks with the researcher and received feedback according to the group they came from. Pre and posttests (immediate and delayed) of accuracy and fluency were conducted using additional picture tasks. Accuracy was measured by calculating the percentage of learners’ (a) error-free clauses and (b) error-free T-units. Fluency was measured by calculating the number of (a) syllables per minute and (b) meaningful syllables per minute. Statistical analyses included (a) two-way repeated measures ANOVAs with feedback type as the between-subject factor and time as the within subject factor, (b) Planned comparisons, which treated the two experimental groups as one group and compared their mean with the mean of the control group, (c) Bonferroni post hoc tests, which examined the pairwise differences, and where needed, (d) paired sample t-tests, which examined each group’s pretest-posttest differences. As for accuracy, planned comparisons showed that videoconferencing CF, irrespective of its explicitness/implicitness, improved accuracy. Further analyses showed that whereas the explicit correction group outperformed the control group on both the immediate and delayed posttests, the recast group did not. However, the explicit feedback group produced a significantly less fluent speech compared to the recast group and the control group. But this was true on the immediate posttest and not on the delayed posttest. Pretest-posttest comparisons further indicated a negative effect for explicit correction but a positive effect for recasts on L2 fluency. The results suggest that (a) while explicit correction assisted accuracy, it negatively influenced fluency, and (b) while implicit correction seemed to assist fluency, it was not as effective as the effect of explicit correction on L2 accuracy. Further analyses indicated that the explicit correction group exhibited a large amount of monitoring behaviour on the immediate posttest, whereas the other two groups did not. The results are explained using an information-processing perspective of language performance and a knowledge proceduralization model of language development. The theoretical, empirical, and pedagogical implications are also discussed. / Graduate

The increasing reliance of people on computers for daily tasks has resulted in a vast number of digital documents. Search engines were once luxury tools for quickly scanning a set of documents but are now quickly becoming the only practical way to navigate through this sea of information. Traditionally, search engine results are based upon a mathematical formula of document relevance to a search phrase. Often, however, what a user deems to be relevant and what a search engine computes as relevant are not the same. User feedback regarding the utility of a search result can be collected in order to refine query results. Additionally, user feedback can be used to identify queries that lack high quality search results. A content author can then further develop existing content or create new content to improve those search results. The most straightforward way of collecting user feedback is to add a graphical user interface component to the search interface that asks the user how much he or she liked the search result. However, if the feedback mechanism requires the user to provide feedback before he or she can progress further with his or her search, the user may become annoyed and provide incorrect feedback values out of spite. Conversely, if the feedback mechanism does not require the user to provide feedback at all then the overall amount of collected feedback will be diminished as many users will not expend the effort required to give feedback. This research focused on the collection of explicit user feedback in both mandatory (a user must give feedback) and voluntary (a user may give feedback) scenarios. The collected data was used to train a set of decision tree classifiers that provided user satisfaction values as a function of implicit user behavior and a set of search terms. The results of our study indicate that a more accurate classifier can be built from explicit data collected in a voluntary scenario. Given a limited search domain, the classification accuracy can be further improved.

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第20031号 / 情博第626号 / 新制||情||109(附属図書館) / 33127 / 京都大学大学院情報学研究科通信情報システム専攻 / (主査)教授 守倉 正博, 教授 原田 博司, 教授 梅野 健 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DGAM

Automated systems for producing product recommendations to users is a relatively new area within the field of machine learning. Matrix factorization techniques have been studied to a large extent on data consisting of explicit feedback such as ratings, but to a lesser extent on implicit feedback data consisting of for example purchases.The aim of this study is to investigate how well matrix factorization techniques perform compared to other techniques when used for producing recommendations based on purchase data. We conducted experiments on data from an online bookstore as well as an online fashion store, by running algorithms processing the data and using evaluation metrics to compare the results. We present results proving that for many types of implicit feedback data, matrix factorization techniques are inferior to various neighborhood- and association rules techniques for producing product recommendations. We also present a variant of a user-based neighborhood recommender system algorithm \textit{(UserNN)}, which in all tests we ran outperformed both the matrix factorization algorithms and the k-nearest neighbors algorithm regarding both accuracy and speed. Depending on what dataset was used, the UserNN achieved a precision approximately 2-22 percentage points higher than those of the matrix factorization algorithms, and 2 percentage points higher than the k-nearest neighbors algorithm. The UserNN also outperformed the other algorithms regarding speed, with time consumptions 3.5-5 less than those of the k-nearest neighbors algorithm, and several orders of magnitude less than those of the matrix factorization algorithms.

Η έλλειψη της δραστηριότητας του χρήστη σχετικά με τα αποτελέσματα της αναζήτησης μέχρι πρόσφατα θεωρείτο ως ένδειξη της δυσαρέσκειας του από την απόδοση ανάκτησης, και συχνά τέτοια αδράνεια χαρακτήριζε την αναζήτηση ως αποτυχημένη (negative search abandonment). Ωστόσο, πρόσφατες μελέτες δείχνουν ότι ορισμένες αναζητήσεις μπορούν να ικανοποιηθούν από το περιεχόμενο των αποτελεσμάτων που παρουσιάζονται στον χρήστη, χωρίς να χρειάζεται να κάνει κλικ σε κάποιο από τα ανακτημένα αποτελέσματα (positive search abandonment), και έτσι τονίζεται η ανάγκη να γίνουν διακρίσεις μεταξύ των επιτυχημένων και αποτυχημένων αναζητήσεων που δεν ακολουθούνται από κλικς. Με αυτή την εργασία προτείνουμε τον σχεδιασμό και την υλοποίηση μιας μεθοδολογίας αξιολόγησης της ικανοποίησης του χρήστη από τα αποτελέσματα αναζητήσεων που δεν ακολουθούνται από επισκέψεις στο περιεχόμενο των δεδομένων ανάκτησης. Για την επίτευξη του στόχου αυτού διενεργήσαμε μελέτη χρηστών που διερευνά τις προθέσεις των χρηστών πίσω από ερωτήματα που δεν ακολουθούνται από επίσκεψη σε κάποιο από τα αποτελέσματα που επέστρεψε η αναζήτηση και εξετάζει τις εργασίες αναζήτησης που μπορούν να ολοκληρωθούν με επιτυχία βασισμένες εξ ολοκλήρου στις πληροφορίες που παρέχονται στη σελίδα με τα αποτελέσματα. Επιπρόσθετα, μελετήθηκαν και υλοποιήθηκαν εργαλεία, QWC Browser, για την καταγραφή της δραστηριότητας του χρήστη με συστήματα ανάκτησης πληροφορίας από τον Παγκόσμιο Ιστό. Στηριζόμενοι στην ευρέως αποδεχόμενη ιδέα της χρήσης της δραστηριότητας του χρήστη ως δείκτη υπονοούμενης αξιολόγησης συσχέτισης (implicit relevance judgments), εξετάσαμε την ύπαρξη σχέση μεταξύ των ρητών δηλώσεων (explicit judgments) ικανοποίησης του χρήστη και μετρικών αξιολόγησης της υπονοούμενης ανατροφοδότησης (implicit measures) του χρήστη. Τέλος, χρησιμοποιήσαμε τεχνικές μοντελοποίησης για την ανάπτυξη μοντέλων πρόβλεψης για την σύλληψη της ικανοποίησης του χρήστη από τις αναζητήσεις που δεν ακολουθούνται από κλικς. / The lack of user activity on search results was until recently perceived as a sign of user dissatisfaction from retrieval performance, often, referring to such inactivity as a failed search (negative search abandonment). However, recent studies suggest that some search tasks can be achieved in the contents of the results displayed without the need to click through them (positive search abandonment); thus they emphasize the need to discriminate between successful and failed searches without follow-up clicks. In this paper we propose to design and implement a methodology for assessing user satisfaction from the results of searches that are not followed by visits to the content of the retrieved results. To achieve this goal we conducted a user study in order to identify the search intentions of queries without follow-up clicks to any of the results returned by the search and identify the search tasks that can be accomplished successfully based entirely on information provided on the results page. Additionally, we developed an instrumented browser, QWC Browser, to collect a variety of measures of user activity after the query submittion. Moreover, we examined whether there is an association between explicit judgments of user satisfaction and implicit measures of user interest in order to understand what implicit measures were most strongly associated with user satisfaction. Finally, we used Bayesian modeling techniques to develop predictive models, to capture user satisfaction from searches that are not followed by clicks to the retrieved results.

隨著全球資訊網蓬勃的發展，可以幫助使用者根據關鍵字搜尋相關資訊的搜尋引擎也已變成使用者不可或缺的工具之一。但對於搜尋引擎生手而言，往往不知道該如何地輸入適當的關鍵字，導致搜尋結果不如預期。如果搜尋引擎可以提供自動查詢修正(Automatic Query Refinement)的功能，將可以有效地幫助生手在網路上找尋到其想要的資訊。因此，如何得知使用者的資訊需求，如何自動化地達到查詢修正，則成為重要的課題之一。本研究利用使用者的隱含回饋(Implicit Feedback)來分析使用者的資訊需求，並探勘過去具有相同資訊需求的使用者經驗，以幫助搜尋引擎生手有效地搜尋網頁，以達到自動查詢修正的目的。 本研究中，在長期情境資訊方面，我們從查詢日誌中去辨別出以往使用者所查詢的關鍵字以及點選過的網頁，接著，在短期情境資訊的部份，我們也記錄下目前使用者的查詢關鍵字以及未點選之網頁。 最後，我們在長期情境中濾除掉搜尋引擎生手的查詢過程，同時探勘出與目前使用者有相似資訊需求的以往經驗使用者之查詢過程關鍵字集合，藉以推薦給目前使用者，完成自動查詢修正。 / World Wide Web search engines can help users to search information by their queries, but novice search engines users usually don’t know how to represent their information need. If search engines can offer query refinement automatically, it will help novice search engine users to satisfy their information need effectively. How to find users’ information need, and how to perform query refinement automatically, have become important research issues. In this thesis, we develop the method to help novice search engine users for satisfying their information need effectively by implicit feedback. Implicit feedback in this research is referring to short-term context and long-term context. In this research, first, long-term context is obtained by identifying each user’s queries and extracting conceptual keywords of clickthrough data in each query session from query logs. Then, we also identify current user’s queries and extract conceptual keywords of non-clickthrough data for short-term context identification. Finally, we filter novice sessions from long-term context, and mine frequent itemsets of past experienced users’ search behavior to suggest the most appropriate new query to current user according to their information need.