Current Social Networking Systems (SNS) such as YouTube are creator-driven systems in which creators create content and users search among available content to find what they want. However, queries from users can be time-sensitive, such as some real-time hot topics, which are difficult to obtain at the very moment due to their timeliness and dynamically changing nature. To address this situation, we quest if the system can directly let a user input a query, match the most relevant users (receivers) based on the query and let the receivers decide whether to respond with the very content. In this way, the user can obtain the most relevant data through highly relevant receivers while reducing the reliance on the system's existing data in the recommendation process as an alternative, a new query-driven SNS paradigm. The main objective is to target the most relevant receivers based on a query. In this case, we propose that by allowing users to provide their very moment ideas as queries, the system searches and ranks well-targeted users based on the semantic content of the query and existing user features. However, the user's feature might be incomplete or missing. To alleviate this issue, we propose a novel two-stage query-driven graph-based user recommender system (QDG) that supports query-to-user matching with dynamic update capabilities. In the first stage, we encode the query and item descriptions into attribute features and perform a similarity search to target the Top-N candidate items. In the second stage, we propose a temporal-based graph neural network (t-GNN), which combines the inductive learning-based GNN with the self-attention-based temporal analysis module to predict the most relevant user-item interaction by simultaneously extracting the existing Spatio-temporal features, where spatial feature represents user's relationship with items and temporal feature represents user's behaviour information. We conducted recommendation simulations on six million users and 150,000 merchants on North America YELP data. Experiments show that the QDG system can accurately target strongly relevant users in the North American population based on the query. To the best of our knowledge, we are the first to propose query-driven SNS and demonstrate its effectiveness in a million-scale Yelp dataset.

Note:

Master of Science / Department of Computing and Information Sciences / Doina Caragea / Over the last decade, the Internet has become an involving medium and user-generated content is continuously growing. Recommender systems that exploit user feedback are widely used in e-commerce and quite necessary for business enhancement. To make use of such user feedback, we propose a new content/collaborative hybrid approach, which is built on top of the recently released hetrec2011-movielens-2k dataset and is an extension of a previously proposed approach, called Weighted Tag Recommender (WTR). The WTR approach makes use of tag information available in hetrec2011-movielens-2k, but it does not use explicit ratings. As opposed to WTR, our modified approach can make use of ratings to capture collaborative filtering and either user-tags, available in the hetrec2011-movielens-2k, or movie keywords retrieved from IMDB, to capture movie content information. We call the two versions of our approach Weighted Tag Rating Recommender (WTRR) and Weighted Keyword Rating Recommender (WKRR), respectively. Movie keywords (which are not user specific) allow us to use all ratings available in hetrec2011-movielens-2k, as WKKR associates the content information from movies with the users, based on their ratings. On the other hand, tags provide more specific information for a user, but limit the usage of the data to the user-movie pairs that have tags (significantly smaller number compared with all pairs that have ratings). Both our keyword and tag representations of users can help alleviate the noise and semantic ambiguity problems inherent in information contributed by users of social networks. Experiments using the WTRR approach on a subset of the dataset (which contains both ratings and tags) show that it slightly outperforms the WKRR approach. However, WKRR can be applied to the whole hetrec2011-movielens-2k dataset and results show that the information from keywords can help build a movie recommender system competitive with other neighborhood based approaches and even with more sophisticated state-of-the-art approaches.

隨着在線電子商務網站，音樂視頻網站和社會性共享推薦網站的迅速發展，網站用戶面臨爆炸性增長的選擇。前所未見的大量選擇導致信息過載問題。信息過載問題是指由於存在數量巨大的信息，用戶不能有效的理解並做出選擇的問題。推薦系統是解決信息過載問題的一個關鍵組成部分。過去數十年，推薦系統技術有了長足的進步。研究重點又基於臨近用戶的方法向基於模型的方法過度。然而，推薦系統仍然不夠成熟完善。在本論文中，我們基於真實生活中遇到的問題提出改善推薦系統的方法。 / 首先，我們提出推薦系統的在線學習算法。傳統推薦系統使用批量式學習算法進行訓練。這些方法容易理解並且容易實現。然而批量式學習算法不能有效應對當今推薦系統所面臨的動態情況。新的用戶和新的物品不斷加入推薦系統。在批量式學習算法框架下，要將這些新用戶新物品納入系統，需要對所有數據進行重新學習。另外，在批量式學習算法的每一個步驟中，我們需要處理所有的數據。在現今推薦系統規模下，這通常是非常耗時的。在線學習算法可以通過對每一個數據點調整模型而解決上述兩個問題。 / 其次，我們深入調查大量推薦系統所作的一個假設。該假設默認推薦系統蒐集的打分數據的分佈和未蒐集到的打分數據的分佈是完全一致的。我們使用在真實推薦系統中蒐集的數據證明這個假設極不可能爲真。使用失數據理論的方法，我們提出一個不基於改假設的模型。我們的模型放棄了這個假設並且能夠得到公正的推薦。 / 再次，我們詳細調研推薦系統中的垃圾用戶問題。垃圾用戶的打分會污染正常用戶的數據並導致正常用戶的體驗受到影響。我們提出使用用戶聲譽系統去記錄用戶的聲譽並利用用戶的聲譽去區分垃圾用戶和正常用戶。我們提出一個聲譽生成系統的框架。許多聲譽生成系統是我們聲譽生成系統框架的一個實做。基於該框架，我們還提出一個基於矩陣分解的用戶聲譽生成系統。該系統擁有出衆的分辨垃圾用戶的能力。 / 最後，我們將基於內容的推薦和協同過濾推薦有機結合以便減輕乃至解決冷啓動問題。冷啓動問題是指推薦系統中關於某個用戶或物品的信息是如此之少以至於系統不能對該用戶或改物品做出有效的推薦。用戶的文字性評價中通常包含大量用戶喜好和物品屬性信息。但用戶的文字性評價通常都被直接棄。我們提出一個同時使用基於內容的方法去處理用戶文字性評價信息，使用協同過濾方法處理用戶打分的整合式推薦模型。我們的模型能有效減輕冷啓動問題的影響並且對黑盒協同過濾算法提供可理解的詞彙標籤。這些標籤有助於幫助推薦系統提供推薦的原因。 / 綜上所述，在本論文中我們解決了推薦系統面臨的實際問題並從各個方面對傳統推薦系統進行改進。大量真實數據上的實驗驗證我們提出方法的有效性和高效性。 / With the rapid development of e-commerce websites, music and video streaming websites and social sharing websites, users are facing an explosion of choices nowadays. The presence of unprecedentedly large amount of choices leads to the information overload problem, which refers to the difficulty a user faces in understanding an issue and making decisions that are caused by the presence of too much information. Recommender systems learn users’ preferences based on past behaviors and make suggestions for them. These systems are the key component to alleviate and solve the information overload problem. Encouraging progress has been achieved in the research of recommender systems from neighborhood-based methods to model-based methods. However, recommender systems employed today are far from perfect. In this thesis, we propose to improve the recommender systems from four perspectives motivated by real life problems. / First and foremost, we develop online algorithms for collaborative filtering methods, which are widely applicable to recommender systems. Traditionally batch-training algorithms are developed for collaborative filtering methods. They enjoy the advantage of easy to understand and simple to implement. However, the batch-training algorithms fail to consider the dynamic scenario where new users and new items join the system constantly. In order to make recommendations for these new users and on these new items, batchtraining algorithms need to re-train the model from scratch. During the training process of batch-training algorithms, all the data have to be processed in each iteration. This is prohibitively slow given the sheer size of users and items faced by a real recommender system. Online learning algorithms can solve both of the problems by updating the model incrementally based on a rating point. / Secondly, we question an assumption made implicitly by most recommender systems. Most existing recommender systems assume that the rating distribution of collected ratings and that of the unobserved ratings are the same. Using data collected from a real life recommender system, we show that this assumption is unlikely to be true. By employing the powerful missing data theory, we develop a model that drops this unrealistic assumption and makes unbiased predictions. / Thirdly we examine the spam problem confronted by recommender systems. The ratings assigned by spam users contaminate the data of a recommender system and lead to deteriorated experience for normal users. We propose to use a reputation estimation system to keep track of users’ reputations and identify spam users based on their reputations. We develop a unified framework for reputation estimation that subsumes a number of existing reputation estimation methods. Based on the framework, we also develop a matrix factorization based method that demonstrates outstanding discrimination ability. / Lastly, we integrate content-based filtering with collaborative filtering to alleviate the cold-start problem. The cold-start problem refers to the situation where a system has too little information concerning a user or an item to make accurate recommendations. With the readily available rich information embedded in review comments, which are generally discarded, we can alleviate the cold-start problem. Additionally, we can tag the black box collaborative filtering algorithm with interpretable tags that help a recommender system to provide reasons on why items are being recommended. / In summary, we solve some of the major problems faced by recommender systems and improve them from various perspectives in this thesis. Extensive experiments on real life large-scale datasets confirm the effectiveness and efficiency of proposed models. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Ling, Guang. / Thesis (Ph.D.) Chinese University of Hong Kong, 2015. / Includes bibliographical references (leaves 169-184). / Abstracts also in Chinese. / Ling, Guang.

This study is a design science research (DSR) project in which a description of the development and evaluation process for several novel technological artifacts will be communicated. Specifically, this study will establish: 1) an ontology of recommender systems issues, 2) an ontology of financial capability goals, and 3) a framework for a Personal Financial Recommender System (PFRS) application designed to improve user financial capability, called FinPathlight. The impetus for the RecSys Issues Ontology is to address a gap in the literature by providing researchers with a comprehensive knowledge classification of the issues and limitations inherent to recommender systems research. The development of a Financial Capability Goals Ontology will contribute domain knowledge classification for technological systems within the domain of finance and serves as a recommendation item knowledgebase for our PFRS. The FinPathlight framework provides the architecture and principles of implementation for a novel, financial-technology (FinTech) PFRS. FinPathlight is designed to improve the financial capability of its users through the recommendation, tracking and assistance with achieving financial capability enhancing goals. This research is notable in that it expands the influence and furthers the relevance of information systems research by providing an explicitly applicable research solution to an area of significant socio-economic importance, financial capability, a heretofore unsolved “wicked problem” (Churchman 1967) domain. In light of current financial conditions, recommender systems research that addresses a problem such as consumer financial capability is a step towards ensuring that information systems research continues to matter and retain its influence and relevance in everyday practice.

Community Question Answering (CQA) websites provide a rapidly growing source of information in many areas. In most CQA implementations there is little effort in directing new questions to the right group of experts. This means that experts are not provided with questions matching their expertise. In this thesis, we propose a framework for automatically routing a newly posted question to the best suited expert. The purpose of this framework is to decrease the waiting time for a personal response. We also investigate the suitability of two statistical topic models for solving this issue and compare these methods against more traditional Information Retrieval approaches. We show that for a dataset constructed from the Stackoverflow website, these topic models outperform other methods in retrieving a set of best experts. We also show that the Segmented Topic Model gives consistently better performance compared to the Latent Dirichlet Allocation Model.

Thesis (M.S.)--Worcester Polytechnic Institute. / Keywords: Data recency problem; Recommender system; Time-based forgetting function; Time-based forgetting strategy; Collaborative filtering system. Includes bibliographical references (p. 73-74).

Many consumer facing companies have large scale data sets that they use to create recommendations for their users. These recommendations are usually based off information the company has on the user and on the item in question. Based on these two sets of features, models are created and tuned to produce the best possible recommendations. A third set of data that exists in most cases is the presence of past interactions a user may have had with other items. The relationships that a model can identify between this information and the other two types of data, we believe, can improve the prediction of how a user may interact with the given item. We propose a method that can inform the model of these relationships during the training phase while only relying on the user and item data during the prediction phase. Using ideas from convolutional neural networks (CNN) and collaborative filtering approaches, our method manipulated the weights in the first layer of our network design in a way that achieves this goal.

Recommender systems have gained a lot of popularity in recent times dueto their application in the wide range of fields. Recommender systems areintended to support users in finding the relevant items based on their interestsand preferences. Recommender algorithms proposed by researchersevolved over time from simple matching recommendations to machine learningalgorithms. One such class of algorithms with increasing focus is oncalled streaming recommender systems, these algorithms treat input data asa stream of events and make recommendations. To evaluate the algorithmsthat work with continuous data streams, stream-based evaluation techniquesare needed. So far, less interest is shown in the research so far on the evaluationof recommender systems in streaming environments.In this thesis, a simple evaluation framework named StreamER that evaluatesrecommender algorithms that work on streaming data is proposed.StreamER is intended for the rapid prototyping and evaluation of incrementalalgorithms. StreamER is designed and implemented using object-orientedarchitecture to make it more flexible and expandable. StreamER can beconfigured via a configuration file, which can configure algorithms, metricsand other properties individually. StreamER has inbuilt support for calculatingaccuracy metrics, namely click-through rate, precision, and recall.The popular-seller and random recommender are two algorithms supportedout of the box with StreamER. Evaluation of StreamER is performed via acombination of hypothesis and manual evaluation. Results have matched theproposed hypothesis, thereby successfully evaluating the proposed frameworkStreamER.

Question and answer (Q&A) forums, as a way for seeking expertise on the Internet, have seen rapid growth in popularity in recent years. The expertise available on most such forums is voluntary, provided by individuals willing to invest their resources for no monetary remuneration. While these forums provide easy access to expertise, the expertise available is often lacking in quality and depth. Two major reasons for this are, the time investment required to participate in such forums, and the lack of a mechanism for identifying experts for specialized questions. We believe a Q&A recommender engine can ameliorate this problem significantly. The two primary contributions of this work are: a) a hierarchical Bayesian model based Q&A recommender, and b) a discussion of metrics to measure the performance of such a Q&A recommender. Two new metrics, responder load and questioner satisfaction, are suggested based on this discussion. These metrics are used to evaluate the performance of the recommender system on datasets harvested from the Yahoo! Answers website. / text