Search

In this paper, we study an important problem of parsing contextual information from user reviews for recommendation purposes. First, we study the ways contextual information is expressed in user reviews and obtain novel insights about it. Among other things, we demonstrate that such type of information tends to appear at the beginning of the review, in longer sentences, in the sentences written in the past tense or using gerund form, and in the sentences referring to some points in time. Second, we propose a novel context parsing method for systematically extracting contextual information from user-generated reviews that rely on the insights obtained in our study. We apply the proposed method to three different Yelp applications (restaurants, hotels, and beauty & spas) and demonstrate that it works well and leads to better recommendation performance than the baseline approaches. Our method systematically extracts more comprehensive sets of relevant contextual variables and corresponding phrases than the baselines. Our analysis also shows the importance of the newly discovered contextual information for recommendation purposes. The obtained results and the proposed method can help to get more comprehensive knowledge about contextual variables in a given application that leads to better recommendations.

In this article, we suggest a novel deep learning recommendation framework that incorporates contextual information into neural collaborative filtering recommendation approaches. Since context is often represented by dynamic and high-dimensional feature space in multiple applications and services, we suggest to model contextual information in various ways for multiple purposes, such as rating prediction, generating top-k recommendations, and classification of users’ feedback. Specifically, based on the suggested framework, we propose three deep context-aware recommendation models based on explicit, unstructured, and structured latent representations of contextual data derived from various contextual dimensions (e.g., time, location, user activity). Offline evaluation on three context-aware datasets confirms that our proposed deep context-aware models surpass state-of-the-art context-aware methods. We also show that utilizing structured latent contexts in the proposed deep recommendation framework achieves significantly better performance than the other context-aware models on all datasets.

Cross domain recommender system constitutes a powerful method to tackle the cold-start and sparsity problem by aggregating and transferring user preferences across multiple category domains. Therefore, it has great potential to improve click-through-rate prediction performance in online commerce platforms having many domains of products. While several cross domain sequential recommendation models have been proposed to leverage information from a source domain to improve CTR predictions in a target domain, they did not take into account bidirectional latent relations of user preferences across source-target domain pairs. As such, they cannot provide enhanced cross-domain CTR predictions for both domains simultaneously. In this paper, we propose a novel approach to cross-domain sequential recommendations based on the dual learning mechanism that simultaneously transfers information between two related domains in an iterative manner until the learning process stabilizes. In particular, the proposed Dual Attentive Sequential Learning (DASL) model consists of two novel components Dual Embedding and Dual Attention, which jointly establish the two-stage learning process: we first construct dual latent embeddings that extract user preferences in both domains simultaneously, and subsequently provide cross-domain recommendations by matching the extracted latent embeddings with candidate items through dual-attention learning mechanism. We conduct extensive offline experiments on three real-world datasets to demonstrate the superiority of our proposed model, which significantly and consistently outperforms several state-of-the-art baselines across all experimental settings. We also conduct an online A/B test at a major video streaming platform Alibaba-Youku, where our proposed model significantly improves business performance over the latest production system in the company., Accepted to KDD21

Cross domain recommender systems have been increasingly valuable for helping consumers identify useful items in different applications. However, existing cross-domain models typically require large number of overlap users, which can be difficult to obtain in some applications. In addition, they did not consider the duality structure of cross-domain recommendation tasks, thus failing to take into account bidirectional latent relations between users and items and achieve optimal recommendation performance. To address these issues, in this paper we propose a novel cross-domain recommendation model based on dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Furthermore, we combine the dual learning method with the metric learning approach, which allows us to significantly reduce the required common user overlap across the two domains and leads to even better cross-domain recommendation performance. We test the proposed model on two large-scale industrial datasets and six domain pairs, demonstrating that it consistently and significantly outperforms all the state-of-the-art baselines. We also show that the proposed model works well with very few overlap users to obtain satisfying recommendation performance comparable to the state-of-the-art baselines that use many overlap users., Comment: Accepted to IEEE TKDE. arXiv admin note: text overlap with arXiv:1910.05189

Unexpected recommender system constitutes an important tool to tackle the problem of filter bubbles and user boredom, which aims at providing unexpected and satisfying recommendations to target users at the same time. Previous unexpected recommendation methods only focus on the straightforward relations between current recommendations and user expectations by modeling unexpectedness in the feature space, thus resulting in the loss of accuracy measures in order to improve unexpectedness performance. Contrast to these prior models, we propose to model unexpectedness in the latent space of user and item embeddings, which allows to capture hidden and complex relations between new recommendations and historic purchases. In addition, we develop a novel Latent Closure (LC) method to construct hybrid utility function and provide unexpected recommendations based on the proposed model. Extensive experiments on three real-world datasets illustrate superiority of our proposed approach over the state-of-the-art unexpected recommendation models, which leads to significant increase in unexpectedness measure without sacrificing any accuracy metric under all experimental settings in this paper., Comment: Accepted at ACM TIST

A geometrical approach to the analysis of a data bank containing information on space structures of proteins is suggested. As a result of the analysis, several relationships concerning possible conformations are found, and also there is obtained a list of polypeptides whose structures differ essentially from typical concepts of the 3D-structure of proteins. Those deviations could indicate possible errors and gaps of the interpreting models in use, as well as some new unknown relations and rules.

This article presents a novel approach to detecting emergency events, such as power outages, that utilizes social media users as “social sensors” for virtual detection of such events. The proposed new method is based on the analysis of the Twitter data that leads to the detection of Twitter discussions about these emergency events. The method described in the article was implemented and deployed by one of the vendors in the context of detecting power outages as a part of their comprehensive social engagement platform. It was also field tested on Twitter users in an industrial setting and performed well during these tests.

In this paper, we propose a novel model RevGAN that automatically generates controllable and personalized user reviews based on the arbitrarily given sentimental and stylistic information. RevGAN utilizes the combination of three novel components, including self-attentive recursive autoencoders, conditional discriminators, and personalized decoders. We test its performance on the several real-world datasets, where our model significantly outperforms state-of-the-art generation models in terms of sentence quality, coherence, personalization and human evaluations. We also empirically show that the generated reviews could not be easily distinguished from the organically produced reviews and that they follow the same statistical linguistics laws., Accepted to EMNLP 2019

Although the broad social and business success of recommender systems has been achieved across several domains, there is still a long way to go in terms of user satisfaction. One of the key dimensions for significant improvement is the concept of unexpectedness . In this article, we propose a method to improve user satisfaction by generating unexpected recommendations based on the utility theory of economics. In particular, we propose a new concept of unexpectedness as recommending to users those items that depart from what they would expect from the system - the consideration set of each user. We define and formalize the concept of unexpectedness and discuss how it differs from the related notions of novelty, serendipity, and diversity. In addition, we suggest several mechanisms for specifying the users’ expectations and propose specific performance metrics to measure the unexpectedness of recommendation lists. We also take into consideration the quality of recommendations using certain utility functions and present an algorithm for providing users with unexpected recommendations of high quality that are hard to discover but fairly match their interests. Finally, we conduct several experiments on “real-world” datasets and compare our recommendation results with other methods. The proposed approach outperforms these baseline methods in terms of unexpectedness and other important metrics, such as coverage, aggregate diversity and dispersion, while avoiding any accuracy loss.

Advances in tourism economics have enabled us to collect massive amounts of travel tour data. If properly analyzed, this data could be a source of rich intelligence for providing real-time decision making and for the provision of travel tour recommendations. However, tour recommendation is quite different from traditional recommendations, because the tourist’s choice is affected directly by the travel costs, which includes both financial and time costs. To that end, in this article, we provide a focused study of cost-aware tour recommendation. Along this line, we first propose two ways to represent user cost preference. One way is to represent user cost preference by a two-dimensional vector. Another way is to consider the uncertainty about the cost that a user can afford and introduce a Gaussian prior to model user cost preference. With these two ways of representing user cost preference, we develop different cost-aware latent factor models by incorporating the cost information into the probabilistic matrix factorization (PMF) model, the logistic probabilistic matrix factorization (LPMF) model, and the maximum margin matrix factorization (MMMF) model, respectively. When applied to real-world travel tour data, all the cost-aware recommendation models consistently outperform existing latent factor models with a significant margin.

Although the area of context-aware recommender systems (CARS) has made a significant progress over the last several years, the problem of comparing various contextual pre-filtering, post-filtering and contextual modeling methods remained fairly unexplored. In this paper, we address this problem and compare several contextual pre-filtering, post-filtering and contextual modeling methods in terms of the accuracy and diversity of their recommendations to determine which methods outperform the others and under which circumstances. To this end, we consider three major factors affecting performance of CARS methods, such as the type of the recommendation task, context granularity and the type of the recommendation data. We show that none of the considered CARS methods uniformly dominates the others across all of these factors and other experimental settings; but that a certain group of contextual modeling methods constitutes a reliable "best bet" when choosing a sound CARS approach since they provide a good balance of accuracy and diversity of contextual recommendations.

This article describes an approach for incorporating externally specified aggregate ratings information into certain types of recommender systems, including two types of collaborating filtering and a hierarchical linear regression model. First, we present a framework for incorporating aggregate rating information and apply this framework to the aforementioned individual rating models. Then we formally show that this additional aggregate rating information provides more accurate recommendations of individual items to individual users. Further, we experimentally confirm this theoretical finding by demonstrating on several datasets that the aggregate rating information indeed leads to better predictions of unknown ratings. We also propose scalable methods for incorporating this aggregate information and test our approaches on large datasets. Finally, we demonstrate that the aggregate rating information can also be used as a solution to the cold start problem of recommender systems.

On the Web, where the search costs are low and the competition is just a mouse click away, it is crucial to segment the customers intelligently in order to offer more targeted and personalized products and services to them. Traditionally, customer segmentation is achieved using statistics-based methods that compute a set of statistics from the customer data and group customers into segments by applying distance-based clustering algorithms in the space of these statistics. In this paper, we present a direct grouping-based approach to computing customer segments that groups customers not based on computed statistics, but in terms of optimally combining transactional data of several customers to build a data mining model of customer behavior for each group. Then, building customer segments becomes a combinatorial optimization problem of finding the best partitioning of the customer base into disjoint groups. This paper shows that finding an optimal customer partition is NP-hard, proposes several suboptimal direct grouping segmentation methods, and empirically compares them among themselves, traditional statistics-based hierarchical and affinity propagation-based segmentation, and one-to-one methods across multiple experimental conditions. It is shown that the best direct grouping method significantly dominates the statistics-based and one-to-one approaches across most of the experimental conditions, while still being computationally tractable. It is also shown that the distribution of the sizes of customer segments generated by the best direct grouping method follows a power law distribution and that microsegmentation provides the best approach to personalization.

The idea that context is important when predicting customer behavior has been maintained by scholars in marketing and data mining. However, no systematic study measuring how much the contextual information really matters in building customer models in personalization applications has been done before. In this paper, we study how important the contextual information is when predicting customer behavior and how to use it when building customer models. It is done by conducting an empirical study across a wide range of experimental conditions. The experimental results show that context does matter when modeling the behavior of individual customers and that it is possible to infer the context from the existing data with reasonable accuracy in certain cases. It is also shown that significant performance improvements can be achieved if the context is "cleverly" modeled, as described in this paper. These findings have significant implications for data miners and marketers. They show that contextual information does matter in personalization and companies have different opportunities to both make context valuable for improving predictive performance of customers' behavior and decreasing the costs of gathering contextual information.

It is crucial to segment customers intelligently in order to offer more targeted and personalized products and services. Traditionally, customer segmentation is achieved using statistics-based methods that compute a set of statistics from the customer data and group customers into segments by applying clustering algorithms. Recent research proposed a direct grouping-based approach that combines customers into segments by optimally combining transactional data of several customers and building a data mining model of customer behavior for each group. This paper proposes a new micro-targeting method that builds predictive models of customer behavior not on the segments of customers but rather on the customer-product groups. This micro-targeting method is more general than the previously considered direct grouping method. We empirically show that it outperforms the direct grouping and statistics-based segmentation methods across multiple experimental conditions and that it generates predominately small-sized segments, thus providing additional support for the micro-targeting approach to personalization.

Data analysts and naive users alike in information-intensive organizations need automated ways to build, analyze, and maintain very large collections of data mining models.

The need to validate large amounts of data with the help of the domain expert arises naturally in many data-intensive applications, including data mining, data stream, and database-related applications. This paper presents a general validation approach that generalizes different expert-driven validation methods developed for specialized validation problems. In particular, we model the validation process as a sequence of validation operators, explore various properties of such sequences, and present theoretical results that provide for better understanding of the validation process. We also address the problem of selecting the best validation sequence among the class of equivalent sequence permutations. We demonstrate that this optimization problem is NP-hard and present two heuristic algorithms for improving validation sequences.

There have been various claims made in the marketing community about the benefits of 1-to-1 marketing versus traditional customer segmentation approaches and how much they can improve understanding of customer behavior. However, few rigorous studies exist that systematically compare these approaches. In this paper, we conducted such a study and compared the predictive performance of aggregate, segmentation, and 1-to-1 marketing approaches across a broad range of experimental settings, such as multiple segmentation levels, multiple real-world marketing data sets, multiple dependent variables, different types of classifiers, different segmentation techniques, and different predictive measures. Our experiments show that both 1-to-1 and segmentation approaches significantly outperform aggregate modeling. Reaffirming anecdotal evidence of the benefits of 1-to-1 marketing, our experiments show that the 1-to-1 approach also dominates the segmentation approach for the frequently transacting customers. However, our experiments also show that segmentation models taken at the best granularity levels dominate 1-to-1 models when modeling customers with little transactional data using effective clustering methods. In addition, the peak performance of segmentation models are reached at the finest granularity levels, skewed towards the 1-to-1 case. This finding adds support for the microsegmentation approach and suggests that 1-to-1 marketing may not always be the best solution

A drawback of traditional data-mining methods is that they do not leverage prior knowledge of users. In prior work, we proposed a method that could discover unexpected patterns in data by using domain knowledge in a systematic manner. In this paper, we present new methods for discovering a minimal set of unexpected patterns by combining the two, independent concepts of minimality and unexpectedness, both of which have been well-studied in the KDD literature. We demonstrate the strengths of this approach experimentally using a case study in a marketing domain.

Previous work on the solution to analytical electronic customer relationship management (eCRM) problems has used either data-mining (DM) or optimization methods, but has not combined the two approaches. By leveraging the strengths of both approaches, the eCRM problems of customer analysis, customer interactions, and the optimization of performance metrics (such as the lifetime value of a customer on the Web) can be better analyzed. In particular, many eCRM problems have been traditionally addressed using DM methods. There are opportunities for optimization to improve these methods, and this paper describes these opportunities. Further, an online appendix (mansci.pubs.informs.org/ecompanion.html) describes how DM methods can help optimization-based approaches. More generally, this paper argues that the reformulation of eCRM problems within this new framework of analysis can result in more powerful analytical approaches.

This paper presents an architecture of the e-Butler service, i.e. a personalized consumer-centric infomediary delivering a broad range of shopping-related services to consumers. In order to provide these services effectively, e-Butler should be "intelligent". This paper describes how a certain level of intelligence can be achieved through the utilization of a broad range of technologies, including data mining, personalization, profiling, recommender systems, data warehousing, Web proxy servers, data monitoring and XML-based technologies, that are integrated into a unique architecture of e-Butler.

In prior work, we provided methods that generate unexpected patterns with respect to managerial intuition by eliciting managers' beliefs about the domain and using these beliefs to seed the search for unexpected patterns in data. Unexpected patterns discovered in this manner represent contradictions or "holes" in domain knowledge which need to be resolved. Given a belief and a set of unexpected patterns, the motivation behind knowledge refinement is that the belief can be made stronger by refining the belief based on the discovered patterns. In this paper we address the problem of incorporating the discovered contradictions into the belief system based on a formal logic approach. Specifically, we present a framework for refinement based on a generic knowledge refinement strategy, describe abstract properties of refinement algorithms that can be used to compare specific instantiations and then describe and compare two specific refinement algorithms based on this framework.

Focusing on the problems of over-specialization and concentration bias, this paper presents a novel probabilistic method for recommending items in the neighborhood-based collaborative filtering framework. For the probabilistic neighborhood selection phase, we use an efficient method for weighted sampling of k neighbors that takes into consideration the similarity levels between the target user (or item) and the candidate neighbors. We conduct an empirical study showing that the proposed method increases the coverage, dispersion, and diversity reinforcement of recommendations by selecting diverse sets of representative neighbors. We also demonstrate that the proposed approach outperforms popular methods in terms of item prediction accuracy, utility-based ranking, and other popular measures, across various experimental settings. This performance improvement is in accordance with ensemble learning theory and the phenomenon of "hubness" in recommender systems.

This paper describes 1:1 Pro system which constructs personal profiles based on customers' transactional histories. The system uses data mining techniques to discover a set of rules describing customers' behavior and supports human experts in validating the rules.

In many e-commerce applications, ranging from dynamic Web content presentation, to personalized ad targeting, to individual recommendations to the customers, it is important to build personalized profiles of individual users from their transactional histories. These profiles constitute models of individual user behavior and can be specified with sets of rules learned from user transactional histories using various data mining techniques. Since many discovered rules can be spurious, irrelevant, or trivial, one of the main problems is how to perform post-analysis of the discovered rules, i.e., how to validate user profiles by separating “good” rules from the “bad.” This validation process should be done with an explicit participation of the human expert. However, complications may arise because there can be very large numbers of rules discovered in the applications that deal with many users, and the expert cannot perform the validation on a rule-by-rule basis in a reasonable period of time. This paper presents a framework for building behavioral profiles of individual users. It also introduces a new approach to expert-driven validation of a very large number of rules pertaining to these users. In particular, it presents several types of validation operators, including rule grouping, filtering, browsing, and redundant rule elimination operators, that allow a human expert validate many individual rules at a time. By iteratively applying such operators, the human expert can validate a significant part of all the initially discovered rules in an acceptable time period. These validation operators were implemented as a part of a one-to-one profiling system. The paper also presents a case study of using this system for validating individual user rules discovered in a marketing application.

The continued growth and proliferation of e-commerce, online services, and Webbased information systems, has led to an explosion of available content and resources in digital form. As a result of user interactions with these resources, tremendous volumes of clickstream, transaction, and user data are collected by organizations in their daily operations. Mining and analyzing such data can help these organizations determine the life time value of clients, design cross marketing strategies across products and services, evaluate the effectiveness of targeted promotional campaigns, and thus provide more effective personalized content to visitors. Automatic personalization is a central technology used in user-adaptive systems to deliver dynamic content, such as textual information, links, advertisements, and product recommendations that are tailored to the needs and interests of a particular user or a segment of users. The process of personalization comprises iterative invocations of the personalization cycle consisting of the stages of understanding the user and building user profiles and models; delivering personalized offerings to the users based on this understanding; measuring impacts of these offerings on the users; and adjusting all the previous stages based on the measured feedback. Data mining is one of the key technologies used in various stages of this personalization process, especially in building user models and delivering personalized offerings to the users. Therefore, it is crucial to develop appropriate data mining methods fitting the corresponding personalization tasks well. This special issue of User Modeling and User-Adapted Interaction explores recent developments and applications of data mining and machine learning techniques in

Despite the growth in the size and complexity of corporate data, the technology for analyzing it has not kept up with the advances in data collection, in that managers mostly need to rely on marketing research and information systems experts to generate the analysis and reports they need. We review some useful approaches here from the computer science and information systems fields for the analysis of large data sets, viz. good data organization and the use of flexible analysis tools, for making the analysis more tractable and user-friendly. These methods are increasingly being adopted by practitioners who are hard-pressed to generate business intelligence from large corporate databases. However, the benefits of these approaches may not be confined only to practitioners, and may apply to academic researchers working with large data sets, as well.

This paper proposes a novel method for estimating unknown ratings and recommendation opportunities and illustrates the practical implementation of the proposed approach by presenting a certain variation of the classical k-NN method in neighborhood-based collaborative filtering systems using weighted percentiles. We conduct an empirical study showing that the proposed method outperforms the standard user-based collaborative filtering approach by a wide margin in terms of item prediction accuracy and utility-based ranking metrics across various experimental settings. We also demonstrate that this performance improvement is not achieved at the expense of other popular performance measures, such as catalog coverage and aggregate diversity. The proposed approach can also be applied to other popular methods for rating estimation.

Numerous proposals for extending the relational data model to incorporate the temporal dimension of data have appeared over the past decade. It has long been known that these proposals have adopted one of two basic approaches to the incorporation of time into the extended relational model. Recent work formally contrasted the expressive power of these two approaches, termed temporally ungrouped and temporally grouped, and demonstrated that the temporally grouped models are more expressive. In the temporally ungrouped models, the temporal dimension is added through the addition of some number of distinguished attributes to the schema of each relation, and each tuple is “stamped” with temporal values for these attributes. By contrast, in temporally grouped models the temporal dimension is added to the types of values that serve as the domain of each ordinary attribute, and the application's schema is left intact. The recent appearance of TSQL2, a temporal extension to the SQL-92 standard based upon the temporally ungrouped paradigm, means that it is likely that commercial DBMS's will be extended to support time in this weaker way. Thus the distinction between these two approaches—and its impact on the day-to-day user of a DBMS—is of increasing relevance to the database practitioner and the database user community. In this paper we address this issue from the practical perspective of such a user. Through a series of example queries and updates, we illustrate the differences between these two approaches and demonstrate that the temporally grouped approach more adequately captures the semantics of historical data.

One of the central problems in the field of knowledge discovery is the development of good measures of interestingness of discovered patterns. Such measures of interestingness are divided into objective measures-those that depend only on the structure of a pattern and the underlying data used in the discovery process, and the subjective measures-those that also depend on the class of users who examine the pattern. The focus of the paper is on studying subjective measures of interestingness. These measures are classified into actionable and unexpected, and the relationship between them is examined. The unexpected measure of interestingness is defined in terms of the belief system that the user has. Interestingness of a pattern is expressed in terms of how it affects the belief system. The paper also discusses how this unexpected measure of interestingness can be used in the discovery process.

A software specification language Templar is defined in this article. The development of the language was guided by the following objectives: requirements specifications written in Templar should have a clear syntax and formal semantics, should be easy for a systems analyst to develop and for an end-user to understand, and it should be easy to map them into a broad range of design specifications. Templar is based on temporal logic and on the Activity-Event-Condition-Activity model of a rule which is an extension of the Event-Condition-Activity model in active databases. The language supports a rich set of modeling primitives, including rules, procedures, temporal logic operators, events, activities, hierarchical decomposition of activities, parallelism, and decisions combined together into a cohesive system.

A high-level simulation language based on temporal logic is described. The language combines a large set of temporal tenses and a rich class of high-level modeling primitives. Also an implementation of the language interpreter is presented. Finally, a real-world case study is described that shows how a programmer can develop structured, reliable, and well-maintainable simulation programs using the language.

Rating sparsity is a critical issue for collaborative filtering. For example, the well-known Netflix Movie rating data contain ratings of only about 1% user-item pairs. One way to address this rating sparsity problem is to develop more effective methods for training rating prediction models. To this end, in this paper, we introduce a collective training paradigm to automatically and effectively augment the training ratings. Essentially, the collective training paradigm builds multiple different Collaborative Filtering (CF) models separately, and augments the training ratings of each CF model by using the partial predictions of other CF models for unknown ratings. Along this line, we develop two algorithms, Bi-CF and Tri-CF, based on collective training. For Bi-CF and Tri-CF, we collectively and iteratively train two and three different CF models via iteratively augmenting training ratings for individual CF model. We also design different criteria to guide the selection of augmented training ratings for Bi-CF and Tri-CF. Finally, the experimental results show that Bi-CF and Tri-CF algorithms can significantly outperform baseline methods, such as neighborhood-based and SVD-based models.

Temporal logic queries on Datalog and negated Datalog programs are studied, and their relationship to Datalog queries on these programs is explored. It is shown that, in general, temporal logic queries have more expressive power than Datalog queries on Datalog and negated Datalog programs. It is also shown that anexistential domain-independent fragment of temporal logic queries has the same expressive power as Datalog queries on negated Datalog programs with inflationary semantics. This means that for finite structures this class of queries has the power of the fixpoint logic.

The increasing availability of large-scale location traces creates unprecedent opportunities to change the paradigm for knowledge discovery in transportation systems. A particularly promising area is to extract energy-efficient transportation patterns (green knowledge), which can be used as guidance for reducing inefficiencies in energy consumption of transportation sectors. However, extracting green knowledge from location traces is not a trivial task. Conventional data analysis tools are usually not customized for handling the massive quantity, complex, dynamic, and distributed nature of location traces. To that end, in this paper, we provide a focused study of extracting energy-efficient transportation patterns from location traces. Specifically, we have the initial focus on a sequence of mobile recommendations. As a case study, we develop a mobile recommender system which has the ability in recommending a sequence of pick-up points for taxi drivers or a sequence of potential parking positions. The goal of this mobile recommendation system is to maximize the probability of business success. Along this line, we provide a Potential Travel Distance (PTD) function for evaluating each candidate sequence. This PTD function possesses a monotone property which can be used to effectively prune the search space. Based on this PTD function, we develop two algorithms, LCP and SkyRoute, for finding the recommended routes. Finally, experimental results show that the proposed system can provide effective mobile sequential recommendation and the knowledge extracted from location traces can be used for coaching drivers and leading to the efficient use of energy.

Previous work on using external aggregate rating information showed that this information can be incorporated in several different types of recommender systems and improves their performance. In this paper, we propose a more general class of methods that combine external aggregate information with individual ratings in a novel way. Unlike the previously proposed methods, one of the defining features of this approach is that it takes into the consideration not only the aggregate average ratings but also the variance of the aggregate distribution of ratings. The methods proposed in this paper estimate unknown ratings by finding an optimal linear combination of individual-level and aggregate-level rating estimators in a form of a hierarchical regression (HR) model that is grounded in the theory of statistics and machine learning.The proposed HR model is general enough so that the standard individual-level recommender systems and naive aggregate methods constitute special cases of this model.We show that for the general HR model, the presence of the aggregate variance, surprisingly, does not significantly improve estimation of unknown ratings vis-a-vis the case when only aggregate average ratings are considered.In the paper, we experimentally show that the optimal linear combination approach significantly dominates all other special cases, including the classical non-aggregated case and our previously studied aggregate methods, and therefore is the method of choice.

This paper describes an approach for incorporating externally specified aggregate ratings information into certain types of collaborative filtering (CF) methods. For a statistical model-based CF approach, we formally showed that this additional aggregated information provides more accurate recommendations of individual items to individual users. Furthermore, theoretical insights gained from the analysis of this model-based method suggested a way to incorporate aggregate information into the heuristic item-based CF method. Both the model-based and the heuristic item-based CF methods were empirically tested on several datasets, and the experiments uniformly confirmed that the aggregate rating information indeed improves CF recommendations. These results also show the power of theory by demonstrating how the insights gained from theoretical developments can shed light on proper selection of good heuristic methods. We also showed the way to introduce scalability and parallelization into the estimation procedure and reported the running time for steps of the estimation procedure for large datasets.

The paper studies the Long Tail problem of recommender systems when many items in the Long Tail have only few ratings, thus making it hard to use them in recommender systems. The approach presented in the paper splits the whole itemset into the head and the tail parts and clusters only the tail items. Then recommendations for the tail items are based on the ratings in these clusters and for the head items on the ratings of individual items. If such partition and clustering are done properly, we show that this reduces the recommendation error rates for the tail items, while maintaining reasonable computational performance.

Recent research showed that including context in customer behavior models improves predictive performance, especially when the unit of analysis is the single customer. Also segmentation has proved to improve the performance of predictive modeling. The research contribution of this work lies in studying interaction effects between segmentation and contextual information. Several experiments were done on a data set coming from an e-commerce application.

The paper presents a method that uses aggregate ratings provided by various segments of users for various categories of items to derive better estimations of unknown individual ratings. This is achieved by converting the aggregate ratings into constraints on the parameters of a rating estimation model presented in the paper. The paper also demonstrates theoretically that these additional constraints reduce rating estimation errors resulting in better rating predictions.

It is crucial to segment customers intelligently in order to offer more targeted and personalized products and services. Traditionally, customer segmentation is achieved using statistics-based methods that compute a set of statistics from the customer data and group customers into segments by applying clustering algorithms. Recent research proposed a direct grouping-based approach that combines customers into segments by optimally combining transactional data of several customers and building a data mining model of customer behavior for each group. This paper proposes a new micro targeting method that builds predictive models of customer behavior not on the segments of customers but rather on the customer-product groups. This micro-targeting method is more general than the previously considered direct grouping method. We empirically show that it significantly outperforms the direct grouping and statistics-based segmentation methods across multiple experimental conditions and that it generates predominately small-sized segments, thus providing additional support for the micro-targeting approach to personalization.

This paper compares customer segmentation, 1-to-1, and aggregate marketing approaches across a broad range of experimental settings, including multiple segmentation levels, marketing datasets, dependent variables, and different types of classifiers, segmentation techniques, and predictive measures. Our experimental results show that, overall, 1-to-1 modeling significantly outperforms the aggregate approach among high-volume customers and is never worse than aggregate approach among low-volume customers. Moreover, the best segmentation techniques tend to outperform 1-to-l modeling among low-volume customers.

In this paper we study market share rules, rules that have a certain market share statistic associated with them. Such rules are particularly relevant for decision making from a business perspective. Motivated by market share rules, in this paper we consider statistical quantitative rules (SQ rules) that are quantitative rules in which the RHS can be any statistic that is computed for the segment satisfying the LHS of the rule. Building on prior work, we present a statistical approach for learning all significant SQ rules, i.e., SQ rules for which a desired statistic lies outside a confidence interval computed for this rule. In particular we show how resampling techniques can be effectively used to learn significant rules. Since our method considers the significance of a large number of rules in parallel, it is susceptible to learning a certain number of "false" rules. To address this, we present a technique that can determine the number of significant SQ rules that can be expected by chance alone, and suggest that this number can be used to determine a "false discovery rate" for the learning procedure. We apply our methods to online consumer purchase data and report the results.

The problem of analyzing microarray data became one of important topics in bioinformatics over the past several years, and different data mining techniques have been proposed for the analysis of such data. In this paper, we propose to use association rule discovery methods for determining associations among expression levels of different genes. One of the main problems related to the discovery of these associations is the scalability issue. Microarrays usually contain very large numbers of genes that are sometimes measured in 10,000s. Therefore, analysis of such data can generate a very large number of associations that can often be measured in millions. The paper addresses this problem by presenting a method that enables biologists to evaluate these very large numbers of discovered association rules during the post-analysis stage of the data mining process. This is achieved by providing several rule evaluation operators, including rule grouping, filtering, browsing, and data inspection operators, that allow biologists to validate multiple individual gane regulation patterns at a time. By iteratively applying these operators, biologists can explore a significant part of all the initially generated rules in an acceptable period of time and thus answer biological questions that are of a particular interest to him or her. To validate our method, we tested our system on the microarray data pertaining to the studies of environmental hazards and their influence of gane expression processes. As a result, we managed to answer several questions that were of interest to the biologists that had collected this data.

In this paper, we present a new data-warehousing-based approach to recommender systems. In particular, we propose to extend traditional two-dimensional user/item recommender systems to support multiple dimensions, as well as comprehensive profiling and hierarchical aggregation (OLAP) capabilities. We also introduce a new recommendation query language RQL that can express complex recommendations taking into account the proposed extensions. We describe how these extensions are integrated into a framework that facilitates more flexible and comprehensive user interactions with recommender systems.

There has been much attention given recently to the task of finding interesting patterns in temporal databases. Since there are so many different approaches to the problem of discovering temporal patterns, we first present a characterization of different discovery tasks and then focus on one task of discovering interesting patterns of events in temporal sequences. Given an (infinite) temporal database or a sequence of events one can, in general, discover an infinite number of temporal patterns in this data. Therefore, it is important to specify some measure of interestingness for discovered patterns and then select only the patterns interesting according to this measure. We present a probabilistic measure of interestingness based on unexpectedness, whereby a pattern P is deemed interesting if the ratio of the actual number of occurrences of P exceeds the expected number of occurrences of P by some user defined threshold. We then make use of a subset of the propositional, linear temporal logic and present an efficient algorithm that discovers unexpected patterns in temporal data. Finally, we apply this algorithm to synthetic data, UNIX operating system calls, and Web logfiles and present the results of these experiments.

In this paper we address some of the concerns that have been expressed regarding the practicality of the temporally-grouped, or history-oriented, data modeling approach. Specifically, we address the concern over the lack of an algebra for this paradigm, by presenting such an algebra. In addition, we examine the semantics of two notions, coalescing and restructuring, from the perspective of a comparison between the temporally grouped and temporally ungrouped paradigms.