Showing papers on "Recommender system published in 1998"

Empirical analysis of predictive algorithms for collaborative filtering

Abstract: Collaborative filtering or recommender systems use a database about user preferences to predict additional topics or products a new user might like. In this paper we describe several algorithms designed for this task, including techniques based on correlation coefficients, vector-based similarity calculations, and statistical Bayesian methods. We compare the predictive accuracy of the various methods in a set of representative problem domains. We use two basic classes of evaluation metrics. The first characterizes accuracy over a set of individual predictions in terms of average absolute deviation. The second estimates the utility of a ranked list of suggested items. This metric uses an estimate of the probability that a user will see a recommendation in an ordered list. Experiments were run for datasets associated with 3 application areas, 4 experimental protocols, and the 2 evaluation metr rics for the various algorithms. Results indicate that for a wide range of conditions, Bayesian networks with decision trees at each node and correlation methods outperform Bayesian-clustering and vector-similarity methods. Between correlation and Bayesian networks, the preferred method depends on the nature of the dataset, nature of the application (ranked versus one-by-one presentation), and the availability of votes with which to make predictions. Other considerations include the size of database, speed of predictions, and learning time.

Learning Collaborative Information Filters

Abstract: Predicting items a user would like on the basis of other users’ ratings for these items has become a well-established strategy adopted by many recommendation services on the Internet. Although this can be seen as a classification problem, algorithms proposed thus far do not draw on results from the machine learning literature. We propose a representation for collaborative filtering tasks that allows the application of virtually any machine learning algorithm. We identify the shortcomings of current collaborative filtering techniques and propose the use of learning algorithms paired with feature extraction techniques that specifically address the limitations of previous approaches. Our best-performing algorithm is based on the singular value decomposition of an initial matrix of user ratings, exploiting latent structure that essentially eliminates the need for users to rate common items in order to become predictors for one another's preferences. We evaluate the proposed algorithm on a large database of user ratings for motion pictures and find that our approach significantly outperforms current collaborative filtering algorithms.

Recommendation as classification: using social and content-based information in recommendation

Abstract: Recommendation systems make suggestions about artifacts to a user. For instance, they may predict whether a user would be interested in seeing a particular movie. Social recomendation methods collect ratings of artifacts from many individuals, and use nearest-neighbor techniques to make recommendations to a user concerning new artifacts. However, these methods do not use the significant amount of other information that is often available about the nature of each artifact - such as cast lists o r movie reviews, for example. This paper presents an inductive learning approach to recommendation that is able to use both ratings information and other forms of information about each artifact in predicting user preferences. We show that our method outperforms an existing social-filtering method in the domain of movie recommendations on a dataset of more than 45,000 movie ratings collected from a community of over 250 users.

Implicit Feedback for Recommender Systems

Abstract: Can implicit feedback substitute for explicit ratings in recommender systems? If so, we could avoid the difficulties associated with gathering explicit ratings from users. How, then, can we capture useful information unobtrusively, and how might we use that information to make recommendations? In this paper we identify three types of implicit feedback and suggest two strategies for using implicit feedback to make recommendations.

Using filtering agents to improve prediction quality in the GroupLens research collaborative filtering system

Abstract: Collaborative filtering systems help address information overload by using the opinions of users in a community to make personal recommendations for documents to each user. Many collaborative filtering systems have few user opinions relative to the large number of documents available. This sparsity problem can reduce the utility of the filtering system by reducing the number of documents for which the system can make recommendations and adversely affecting the quality of recommendations. This paper defines and implements a model for integrating content-based ratings into a collaborative filtering system. The filterbot model allows collaborative filtering systems to address sparsity by tapping the strength of content filtering techniques. We identify and evaluate metrics for assessing the effectiveness of filterbots specifically, and filtering system enhancements in general. Finally, we experimentally validate the filterbot approach by showing that even simple filterbots such as spell checking can increase the utility for users of sparsely populated collaborative filtering systems.

Collecting user access patterns for building user profiles and collaborative filtering

Abstract: The paper proposes a new learning mechanism to extract user preferences transparently for a World Wide Web recommender system. The general idea is that we use the entropy of the page being accessed to determine its interestingness based on its occurrence probability following a sequence of pages accessed by the user. The probability distribution of the pages is obtained by collecting the access patterns of users navigating on the Web. A finite context-model is used to represent the usage information. Based on our proposed model, we have developed an autonomous agent, named ProfBuilder, that works as an online recommender system for a Web site. ProfBuilder uses the usage information as a base for content-based and collaborative filtering.

Recommendation systems: a probabilistic analysis

Abstract: A recommendation system tracks past actions of a group of users to make recommendations to individual members of the group. The growth of computer-mediated marketing and commerce has led to increased interest in such systems. We introduce a simple analytical framework for recommendation systems, including a basis for defining the utility of such a system. We perform probabilistic analyses of algorithmic methods within this framework. These analyses yield insights into how much utility can be derived from the memory of past actions and on how this memory can be exploited.

Exploring Versus Exploiting when Learning User Models for Text Recommendation

Abstract: The text recommendation task involves delivering sets of documents to users on the basis of user models. These models are improved over time, given feedback on the delivered documents. When selecting documents to recommend, a system faces an instance of the exploration/exploitation tradeoff: whether to deliver documents about which there is little certainty, or those which are known to match the user model learned so far. In this paper, a simulation is constructed to investigate the effects of this tradeoff on the rate of learning user models, and the resulting compositions of the sets of recommended documents, in particular World-Wide Web pages. Document selection strategies are developed which correspond to different points along the tradeoff. Using an exploitative strategy, our results show that simple preference functions can successfully be learned using a vector-space representation of a user model in conjunction with a gradient descent algorithm, but that increasingly complex preference functions lead to a slowing down of the learning process. Exploratory strategies are shown to increase the rate of user model acquisition at the expense of presenting users with suboptimal recommendations; in addition they adapt to user preference changes more rapidly than exploitative strategies. These simulated tests suggest an implementation for a simple control that is exposed to users, allowing them to vary a system‘s document selection behavior depending on individual circumstances.

Comparing feature-based and clique-based user models for movie selection

Abstract: The huge amount of information available in the currently evolving world wide information infrastructure at any one time can easily overwhelm end-users. One way to address the information explosion is to use an “information filtering agent” which can select information according to the interest and/or need of an end-user. However, at present few such information filtering agents exist. In this study, we evaluate the use of feature-based approaches to user modcling with the purpose of creating a filtering agent for the video-on-demand application. We evaluate several feature and clique-based models for 10 voluntary subjects who provided ratings for the movies. Our preliminary results suggest that feature-based selection can be a useful tool to recommend movies according to the taste of the user and can be as effective as a movie rating expert. We compare our feature-based approach with a clique-based approach, which has advantages whcrc information from other users is available.

Book Recommending Using Text Categorization with Extracted Information

Abstract: Content-based recommender systems suggest documents, items, and services to users based on learning a prole of the user from rated examples containing information about the given items. Text categorization methods are very useful for this task but generally rely on unstructured text. We have developed a bookrecommending system that utilizes semi-structured information about items gathered from the web using simple information extraction techniques. Initial experimental results demonstrate that this approach can produce fairly accurate recommendations.

Recommender Systems for TV

Abstract: We present research questions associated with recommender systems for TV and an example of such a

Learning to surf: multiagent systems for adaptive web page recommendation

Abstract: Imagine a newspaper personalized for your tastes. Instead of a selection of articles chosen for a general audience by a human editor, a software agent picks items just for you, covering your particular topics of interest. Since there are no journalists at its disposal, the agent searches the Web for appropriate articles. Over time, it uses your feedback on recommended articles to build a model of your interests. This thesis investigates the design of "recommender systems" which create such personalized newspapers. Two research issues motivate this work and distinguish it from approaches usually taken by information retrieval or machine learning researchers. First, a recommender system will have many users, with overlapping interests. How can this be exploited? Second, each edition of a personalized newspaper consists of a small set of articles. Techniques for deciding on the relevance of individual articles are well known, but how is the composition of the set determined? One of the primary contributions of this research is an implemented architecture linking populations of adaptive software agents. Common interests among its users are used both to increase efficiency and scalability, and to improve the quality of recommendations. A novel interface infers document preferences by monitoring user drag-and-drop actions, and affords control over the composition of sets of recommendations. Results are presented from a variety of experiments: user tests measuring learning performance, simulation studies isolating particular tradeoffs, and usability tests investigating interaction designs.

An Interface for Learning Multi-topic User Profiles from Implicit Feedback

Abstract: A text recommender system recommends sets of documents for individual users on the basis of user models, which are incrementally constructed given feedback on previous recommendations Users are reluctant to take the time to provide such feedback explicitly One of the contributions of this research is an interface design for a recommender system which infers document preferences by monitoring users’ actions A second problem for recommender systems is determining the composition of a set of recommendations, especially when users have many interests The interface presented provides a mechanism for users to define multiple topics of interest and control the proportions between them Observations from initial usability tests are encouraging--they demonstrate the system successfully learning multi-topic user profiles using only the implicit feedback of users’ clicking and dragand-drop actions

Recommender Systems for Problem Solving Environments

Abstract: Traditionally, recommender systems have been studied in domains that focus on harnessing distributed information resources, collaborative filtering, information aggregation, social schemes for decision making and user interfaces. In this paper, we present recommender systems in a different context -- primarily as systems that select (scientific) software appropriate to a user’s needs. This application is motivated by the wide acceptance of problem solving environments (PSEs) which are high level environments for doing computational science. We give an overview of the domain, argue the need for recommender systems and describe our work in this area. The research issues in this discipline are also highlighted.

A new generation method of a user profile for information filtering on the Internet

Abstract: It is very difficult for search engines to find adequate information, which reflects user's interests that are changing with time on Internet. We need various learning techniques to know the user's interests with only given simple query, and to make adaptive Information Retrieval Agents. In this paper, we propose a learning mechanism that can be used in making it easier to construct the user profile from accepted relevant documents' terms. These documents are proposed by the user's relevance feedback and we select key-terms in the documents for making profile. With using this profile we can filter the retrieved information and represent more relevant information to user. This approach has applied to an autonomous information gathering and filtering system, named IGIMA (Intelligent Information Gathering and Filtering System based on Multi-Agent).

A Hyperlink-Based Recommender System Written in Sqeal.

Abstract: 1. ABSTRACT Human beings, not machines, are the ultimate experts for information retrieval tasks, including recommender systems. Consequently, computers are most useful when they combine information about people’s judgments. Collaborative filtering systems make use of this observation by having users explicitly rate items, such as Web pages, with the system making recommendations to other users based on overlapping areas of interest. A disadvantage of collaborative filtering, at least as currently implemented, is that it depends on users’ explicitly entering data, which can be inconvenient and time-consuming. We describe the design, implementation, and performance of a recommender system that works by mining publiclyavailable hyperlinks on the Web, producing results competitive with the best text-based system. We also demonstrate the utility of the Squeal language for structure-based Web queries.

Collaborative Filtering for Web Marketing Efforts

Abstract: Recommender systems can improve consumer response to ecommerce sites. Large commercial sites make extraordinary resource demands on systems and databases. This paper describes a case study involving Columbia House, a large consumer direct marketing firm. Columbia House required rapid response time with high traffic, good recommendations at site-opening, and the ability to recommend new titles as they became available. LikeMinds developed a parallel collaborative filtering recommender that provided nearly linear speedup, making recommendations in less than 30 milliseconds on a single processor. A technique called “composite archetypes” helped seed the database with information from legacy transaction databases, making good recommendations at the time of the site opening. Another technique called “objective archetypes” allowed new titles to be recommended using only categorical information. Introduction Collaborative filtering is a real-time personalization technique. It leverages similarities between people to make recommendations, and can be more accurate than mechanical techniques, such as rule-based systems or Bayesian methods. LikeMinds uses collaborative filtering technology to improve marketing efforts for direct marketers: accurate personalization tends to drive up revenues on commercial web sites. LikeMinds developer John Hey created the first general collaborative filtering system in 1987. It was originally deployed in video recommender kiosks around Boston. Since that time, web commercialization has placed increasing demands on the technique. Commercial Needs Columbia House was one of the first large consumer retailers to move to the web. Columbia House markets music, videos and CDROMs directly to consumers as a loyalty club, originally through a direct mail marketing channel. Music customers receive 12 titles free for signing up, promising to purchase 6 more titles at regular price. Columbia House has historically used rule-based personalization in its business, hiring a small army of analysts to assign each new album to one or more listening preferences. The first Columbia House web site used this personalization method. However, with broad categories, such as "alternative rock," rule-based personalization was not very personal. Columbia House decided to create a new site that would make more individualized recommendations to visitors. It chose LikeMinds collaborative filtering technology for this effort. The Columbia House site required real-time performance under heavy usage. Columbia House has the second highest number of ecommerce transactions of all web sites, and offers thousands of products. It had an existing relational database that provided customer and product information, and wanted to store ratings and intermediate values in its database. For the first time, a major consumer products company wanted to integrate collaborative filtering into every aspect of a site, making performance a critical issue. Whenever a customer searched for albums, the site would state which albums were recommended. Nearly every page view demanded personalization. The recommended albums were to be different for each customer: segmentation, which increases speed at the expense of accuracy, could not be used to speed the process. From: AAAI Technical Report WS-98-08. Compilation copyright © 1998, AAAI (www.aaai.org). All rights reserved.

Recommending TV Programs: How far can we get at zero user effort?

Abstract: RECOMMENDATION REQUIRES EFFORT Collaborative filtering methods have been applied to a number of domains like books, videos, audio CDs and Usenet news. These systems require some effort on the part of users before they can generate recommendations (see for example [5] on the “cold-start” problem). How much effort they require depends on domain and application. Some recommender systems for books or videos require rating a specific number of items before they generate the first prediction. If users can save money by not buying the wrong books and videos this effort might pay off. The Usenet news filter GroupLens [2] gathers ratings while users read articles so the process of building the profile and getting recommendations is interleaved. This strategy might lower the threshold for getting started. How much effort a user is willing to make depends not only on the domain but also on the personal value of the recommendation. Users who depend on the information in a newsgroup might make every required effort, but occasional users with a less serious information interest might not. To summarize: The higher the required effort the more potential users will quit.

Visual querying and explanation of recommendations from collaborative filtering systems

Abstract: INTRODUCTION Collaborative filtering is a technique that makes use of kuowledge from other users to find useful information by computing similarity of the users based on their rating patterns [2]. Although this technique can deal with a user’s subjective “taste” for data such as movies and music, one of its problems is that a user’s taste is diverse and changing; the filter might fit only a portion of the user’s taste or fail to satisfy her or his temporary needs. We claim that the system shouid explain how filtered items match the user’s taste and give users control so that they can explore the information space to find what they want.

Text Categorization Through Probabilistic Learning: Applications to Recommender Systems

Abstract: With the growth of the World Wide Web, recommender systems have received an increasing amount of attention. Many recommender systems in use today are based on col laborative ltering. This pro ject has focused on LIBRA, a content-based book recommending system. By utilizing text categorization methods and the information available for each book, the system determines a user prole which is used as the basis of recommendations made to the user. Instead of the bagof-words approach used in many other statistical text categorization approaches, LIBRA parses each text sample into a semi-structured representation. We have used standard Machine Learning techniques to analyze the performance of several algorithms on this learning task. In addition, we analyze the utility of several methods of feature construction and selection (i.e. methods of choosing the representation of an item that the learning algorithm actually uses). After analyzing the system we conclude that good recommendations are produced after a relatively small number of training examples. We also conclude that the feature selection method tested does not improve the performance of these algorithms in any systematic way, though the results indicate other feature selection methods may prove useful. Feature construction, however, while not providing a large increase in performance with the particular construction methods used here, holds promise of providing performance improvements for the algorithms investigated. This text assumes only minor familiarity with concepts of articial intelligence and should be readable by the upper division computer science undergraduate familiar with basic concepts of probability theory and set theory.

Towards a Framework for Building Collaborative Information Searching Systems

Abstract: If the World Wide Web (the web for short), should become the world wide digital library, not only effective and effecient information searching techniques are needed, but also an adequate collaboration support that enable people to cooperate and collaborate in locating relevant information just as they do in physical libraries. Collaboration support is required during information searching as well as for sharing results of previous searching process. Collaborative information searching (CIS) can be either direct or indirect. In direct collaboration, people communicate directly, in synchronous or asynchronous manner, in order to show one another where to go to find a given information or simply to send to one another the required information. In indirect collaboration information gathered from previous information searching process conducted a user are used to help other users in their searching activities. Recommender systems are an example of indirect CIS systems. In this work we address the problem of providing a framework that facilitates the design and the implementation of various CIS applications. An overview of this framework, called, Broadway*Tools is presented in the next section.

Interactive Interface Agents as Recommender Systems

Abstract: This paper presents a model for more interactive interface agents. Using learning interface agents is one strategy for designing recommender systems. The more interactive style of agents presented in this paper aims to increase the trust and understanding between the user and the agent, by allowing the agent to solicit further input from the user under certain conditions. We illustrate our design for more interactive interface agents by including some examples in the domain of electronic mail. We then discuss why the model is also applicable to designing recommender systems, in general.

Content + Collaboration - Recommendation

Abstract: Building recommender systems, that filters and suggests relevant and personalized information to its user, has become a practical challenge for researchers in AI. On the other hand, the relevance of such information is a userdependent notion, defmed within the scope or context of a particular domain or topic. Previous work, mainly in IR, focuses on the analysis of the content by means of keyword-based metrics. Some recent algorithms apply social or collaborative information filtering to improve the task of retrieving relevant information, and for refining each agent’s particular knowledge. In this paper, we combine both approaches developing a new content-based filtering technique for learning up-to-date users’ profile that serves as basis for a novel collaborative information-filtering algorithm. We demonstrate our approach through a system called RAAP (Research Assistant Agent Projec0 implemented to support collaborative research by classifying domain specific information, retrieved from the Web, and recommending these "bookmarks" to other researcher with similar research interests.

A web based management of references

Abstract: During the evolution of research from the beginning of a project to the end a large amount of information is accumulated from books, journals, articles, manuals and the internet. Managing all this information is a complex and crucial part, especially since it is important to keep track of and acknowledge the works of those who helped to formulate their ideas and also to show that their work is original. Often these references are made available to groups of researchers from websites. Unfortunately on-line bibliography databases often contain hundreds if not thousands of records. When a user visits these sites they are often overwhelmed with inappropriate information. This means Academic Staff have to consciously search through the references. Academic Staff would prefer to have documents and references automatically recommended to them by the application. The main focus of this work is to design and implement a suitable Recommender System to filter and promote the information that users really wanted, helping like researchers to collaborate together more effectively to make a further contribution to world knowledge. A fully functional reference management system that incorporated both BibTex upload, download, and restricted security facilities has been implemented. By employing recommendation techniques that had been successfully used by commercial and publication websites like Amazon and CiteSeer we designed and implemented several suggestion systems, based on non personalized, and personalized short-term and long term user interests. The Web Based Reference Management System suggests references to the user through the Statistical Summarization, average user (ratings, reviews, and alternate) recommendations. The final recommendation system implemented provides recommendations for references based on a learned user profile, making use of both hotList and coldList user profiles using a form of incremental Hebbian and Anti Hebbian learning rule to incrementally adapt a feature vector providing a fully functional Content Based Filtering System.

Putting Recommender Systems to Work for Organizations

Abstract: The issue we focus on here is how to make recommender systems work in organizations and for organizations. Moving from the Internet to Intranets requires shifting the primary focus from sharing recommendations to sharing knowledge and from community-building to community support. Moving recommender systems from the Internet onto Intranets also means turning "leisure-ware" into groupware, creating both new challenges and new opportunities.

Mining the Web's Hyperlinks for Recommendations

Abstract: Human beings, not machines, are the ultimate experts for information retrieval tasks, including recommender systems. Consequently, computers are most useful when they combine information about people’s judgments. Collaborative filtering systems make use of this observation by having users explicitly rate items, such as Web pages, with the system making recommendations to other users based on overlapping areas of interest. A disadvantage of collaborative filtering, at least as currently implemented, is that it depends on users’ explicitly entering data, which can be inconvenient and timeconsuming. We describe a recommender system, which we call ParaSite, that works by mining publicly-available hyperlinks on the Web, producing results competitive with the best text-based system.

A data warehousing approach for building recommender systems

Abstract: A data warehousing approach for recommender sys- tems is proposed. We sketch an architecture for integrated OLAP and data mining in data waxe- housing environments, and argue why this archi- tecture can be extended for building recommender systems. Since producing recommendations can be considered as conceptual query answering, the re- lationship between conceptual query answering and intensional answers is also briefly examined.

Using a Recommender System and Hyperwave Attributes To Augment an Electronic Resource Library.

Abstract: There has been increasing interest over the past few years in systems that help users exchange recommendations about World Wide Web documents. Programs have ranged from those that rely totally on user pre-selection, to others that are based on artificial intelligence. This paper proposes a system that falls between these two extremes, providing distinct advantages over other existing systems since it utilizes the structure and attributes associated with Hyperwave documents. This recommender system, which also ensures integrity of external links through its underlying server environment, can enhance the knowledge base of learning associations and other intranet groups. Following an introduction to the context of the project and user participation, the paper briefly describes the features of the Hyperwave server, a document management system for large quantities of multimedia data that can be spread over multiple remote servers. The Simple Recommender System (SRS), prototyped at the University of Auckland (New Zealand), is then detailed, including client-server arrangements and features that support making and receiving recommendations. The issue of maintaining referential integrity within a library of recommendations is considered. (Author/AEF) ******************************************************************************** * Reproductions supplied by EDRS are the best that can be made * * from the original document. * ******************************************************************************** Using a Recommender System and Hyperwave Attributes to Augment an Electronic Resource Library U.S. DEPARTMENT OF EDUCATION Office of Educational Research and Irnprovernent EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC) 13 This document has been reproduced as received from the person or organization originating it. O Minor changes have been made to improve reproduction quality. Points of view or opinions stated in this document do not necessarily represent official OERI position or policy. B. Fenn Aotea Interactive Media Aotearoa New Zealand barry@aotea.co.nz J. Lennon Hyper Media Unit, Computer Science Department The University of Auckland, New Zealand jennifer@cs.auckland.ac.nz "PERMISSION TO REPRODUCE THIS MATERIAL HAS BEEN GRANTED BY G.H. Marks TO THE EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC)." Abstract: There has been increasing interest over the past few years in systems that help users exchange recommendations about web documents. Programs have ranged from those that rely totally on user pre-selection, to others that are based on artificial intelligence. We propose a system that falls between these two extremes, which has distinct advantages over other existing systems since it utilises the structure and attributes associated with Hyperwave documents. This recommender system, which also ensures integrity of external links through its underlying server environment, can enhance the knowledge base of learning associations and other intranet groups. There has been increasing interest over the past few years in systems that help users exchange recommendations about web documents. Programs have ranged from those that rely totally on user pre-selection, to others that are based on artificial intelligence. We propose a system that falls between these two extremes, which has distinct advantages over other existing systems since it utilises the structure and attributes associated with Hyperwave documents. This recommender system, which also ensures integrity of external links through its underlying server environment, can enhance the knowledge base of learning associations and other intranet groups.