K. Lang. Newsweeder: Learning to filter news. In ICML 95, pages 331--339, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....frequency of the word s appearance in the Stan Matwin Computer Science Dept. University of Ottawa Ottawa, ON K1N 6N5 (Canada) stan ca. uot t awa. ca text. This text representation, referred to as the bagof words, is used in most typical approaches to text classification (for recent work see [Lang 95] oachims 97] and [Koller Sahami 97] In these approaches, no linguistic processing (other than a stop list of most frequent words) is applied to the original text. This paper explores the hypothesis that incorporating linguistic knowledge into text representation can lead to improvements ....

K. Lang. NewsWeeder: Learning to Filter News. In Proc. ICML-95, 331-336, 1995,

....Recommender systems have emerged as an important application area and have been the focus of considerable recent academic and commercial interest. The 1997 special issue of the Communications of the ACM [13] contains some key papers. Other important contributions include [2] 4] 7] 12] 15] [8], 1] 11] and [14] In addition, many online retailers are using this technology to recommend new items to their customers, based on what they have bought in the past. Currently, most recommender systems are either content based or collaborative, depending on the type of information that the ....

Ken Lang. Newsweeder: Learning to filter news. In Proceedings of the Twelfth International Conference on Machine Learning, 1997.

....many years, but the recent flood of online text has increased the interest in and applications for text categorization. Internet related classification research has addressed the problem of learning to collect interesting postings to electronic discussion groups based on a user s predilections [ Lang, 1995 ] automatically classifying web pages by content [ Craven et al. 1998 ] and suggesting web pages to a user based on his or her expressed preferences [ Pazzani et al. 1996 ] We focus here on a restricted version of the general classification problem namely, we imagine documents have ....

....amounts and numbers to canonical forms, map all words to uppercase, and remove words occurring twice or less. Table 1 summarizes some salient characteristics of these datasets. 20 Newsgroups: This is a collection of about 20, 000 documents, culled from postings to 20 Usenet discussion groups [ Lang, 1995 ] The documents are approximately evenly distributed among the 20 labels. Four universities: This (misnamed) dataset contains web pages gathered from a large number of university computer science departments [ Craven et al. 1998 ] The pages were manually classified into the categories ....

K. Lang. Newsweeder: Learning to filter news. In Proceedings of the 12th International Conference on Machine Learning, pages 331--339, 1995.

....of such technology that are worth pointing out. We present these applications to show that our work on improved methods for document classification hold promise not only in the context of our system SONIA, but also in a variety of other stand alone applications. Lang s NewsWeeder system [101] shows one application of classification to Usenet Newsgroup article routing. After learning from user judgments of previously read news articles, the system scours through a wide variety of newsgroups collecting other relevant articles for the user. Similarly, Balabanovic has developed a system, ....

Lang, K. NewsWeeder: Learning to filter news. In Machine Learning: Proceedings of the Twelfth International Conference (1995), Morgan Kaufmann.

....filtering of relevant information and presenting organized information (the summary) to the user. Such automated methods, commonly referred to as intelligent information retrieval are used to locate and retrieve information with respect to a user s individual preferences [Balabanovic, 1998; Lang, 1995; M.Pazzani and D.Billsus, 1997; Maglio and Barett, 1997; Thomas and Fischer, 1996 ] Also, efforts to rank sort information based on user preferences, which are often conflicting, has generated interest over the past few years [Schafer et al. 1999a ] One key area of research to achieve this ....

....of the user for movie contents based on previously rated movies by a user. We have accommodated such keyword learning process into our system recommendation by adding a new dimension to our ranking engine. We use a well known algorithm to classify text, based on the naive Bayes classifier [Lang, 1995; Mitchell, 1997; M.Pazzani and D.Billsus, 1997 ] The basic idea is to observe the frequencies of occurrences of words in the synopsis of rated movies. Then the rating of a new movie is based on the content of its synopsis. The more a movie synopsis contains words that occur more frequently in ....

K Lang. Newsweeder: Learning to filter news. In Proceedings of the 12th International conference on machine learning, pages 331--339. Morgan Kaufmann, San Fransisco, 1995.

....of alternatives often makes a wise choice impossible without some intelligent computational assistance. In response to this need, there have been increased efforts to design and implement intelligent aides for filtering web sites (e.g. Pazzani, Muramatsu, Billsus (1996) news stories (e.g. Lang (1995)) TV listing (Smyth and Cotter, 1999) and other information sources. A related line of research and development has led to recommendation systems (e.g. Burke, Hammond, and Young (1996) Resnick and Varian (1997) Burke (1999) which are not limited to filtering information but can be used for ....

....from the one commonly assumed in recommendation systems, which emphasizes improving accuracy or related measures like precision and recall. We want to improve both the subjective quality of the results and the dialogue process. While some adaptive recommendation systems (e.g. Pazzani et.al. 1996, Lang 1995, Linden, Hanks and Lesh 1997, Smyth and Cotter 1999) require the user to provide direct feedback to generate the user model, our basic approach is to derive the preferences of the users from their interactions with the system. To efficiently provide the users with the solution that matches their ....

Lang K, `NEWSWEEDER: Learning to filter news', in `Proceedings of the Twelfth Conference on Machne Learning', pp.331-339, Lake Tahoe, CA, Morgan Kaufmann, 1995.

....the user s interest in a Web page. We report the results of a pilot study. Introduction Much research has been devoted to the task of developing intelligent agents that can learn a user s interests (a profile ) and find information in the World Wide Web (WWW) based on such profiles (e.g. Lang, 1995; Pazzani, Muramatsu, Billsus, 1996; Joachims, Freitag Mitchell, 1997; Shavlik, Calcari, Eliassi Rad Solock, 1999) Given a particular web page or hyperlink and a particular user, the agent s task is to predict the user s interest lin that page or hyperlink. If the agent predicts that the ....

....currently evaluating other learning algorithms for this task. We base our representation of the web page on the bagof words representation (Salton, 1991) this representation is compatible with neural networks and empirically has proven quite successful in information retrieval (Salton, 1991; Lang, 1995; Pazzani et al. 1996) Actually, our agent uses an enhanced version of the basic bag of words representation. We equip our agent with the ability to handle key phrases, which are simply phrases the agent is to look for as opposed to single words. Finally, we address the issue of acquiring ....

K. Lang (1995). NewsWeeder: Learning to Filter News, Proc. ICML-95, pp. 331-339.

....many years, but the recent flood of online text has increased the interest in and applications for text categorization. Internet related classification research has addressed the problem of learning to collect interesting postings to electronic discussion groups based on a user s predilections [ Lang, 1995 ] automatically classifying web pages by content [ Craven et al. 1998 ] and suggesting web pages to a user based on his or her expressed preferences [ Pazzani et al. 1996 ] We focus here on a restricted version of the general classification problem namely, we imagine documents have ....

....amounts and numbers to canonical forms, map all words to uppercase, and remove words occurring twice or less. Table 1 summarizes some salient characteristics of these datasets. ffl 20 Newsgroups: This is a collection of about 20; 000 documents, culled from postings to 20 Usenet discussion groups [ Lang, 1995 ] The documents are approximately evenly distributed among the 20 labels. ffl Four universities: This (misnamed) dataset contains web pages gathered from a large number of university computer science departments [ Craven et al. 1998 ] The pages were manually classified into the categories ....

K. Lang. Newsweeder: Learning to filter news. In Proceedings of the 12th International Conference on Machine Learning, pages 331--339, 1995.

....URL:www.filmfinder.com) which recommends films that users might like. Content based and collaborative filtering are not mutually exclusive. A combination of these methods is used in WiseWire 5 . This is used to recommend news stories and web pages. The interface derives from Lang s NewsWeeder [11]. Content based filtering learns to predict topics which the user has previously taken an interest in, while the collaborative filtering can predict topics which the user hasn t yet seen but may be of interest. 5.3 Generative Interfaces A generative interface creates new data on the basis of ....

K. Lang. NewsWeeder: Learning to Filter News. In Proceedings of the Twelfth International Conference on Machine Learning, pages 331--339. Morgan Kaufmann: Lake Tohoe, CA, 1995.

....sheer amount of information readily available today has created novel challenges. Numerous intelligent information agents software tools that provide personalized assistance for users navigating large information spaces have been described in the literature and deployed on the World Wide Web [5, 8]. However, the need for intelligent information agents is not limited to web based applications, as we are currently witnessing an increasing trend towards ubiquitous information access . Different types of wireless information devices, designed to tap into the Internet s vast information ....

Lang, K. NewsWeeder: learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning (Lake Tahoe CA, 1995), 331--339.

....when deciding whether to recommend that document to the user, which biases it toward documents that are similar to ones the user has previously ranked highly. Although recommending Web pages is a common application of adaptive information filtering, other uses are also possible. Another system, Lang s (1995) NewsWeeder, recommends news stories to readers, again using the words in each story to predict whether the user will find it interesting. Another popular task involves sorting and prioritizing electronic mail, typically using words that occur in the message headers and body (e.g. Boone, 1998) A ....

Lang, K. (1995). NewsWeeder: Learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning (pp. 331--339). Lake Tahoe, CA: Morgan Kaufmann.

....contains nine words a, href, http, golgi, harvard, edu, biopages, all, and html. All words are converted to upper case. Not all words that appear in an HTML document are used as features. We use an information based approach, similar to that used by an early version of the NewsWeeder program (Lang, 1995) to determine which words to use as features. Intuitively, one would like words that occur frequently in pages on the hotlist, but infrequently on pages on the coldlist (or vice versa) This is accomplished by finding the mutual information (e.g. Quinlan, 1984) between the presence or absence of ....

....prevent longer documents from having a better chance of retrieval, the weighted term vectors are normalized to unit length. In Syskill Webert we use the average of the TF IDF vectors of all examples of one class in order to get a prototype vector for the class (similar to the NewsWeeder program, Lang, 1995). To determine the most likely class of an example we convert it to a TF IDF vector and then apply the cosine similarity measure to the example vector and each class prototype. An example is assigned to the class that has the smallest angle between the TF IDF vector of the example and the class ....

Lang, K. (1995). NewsWeeder: Learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning. Lake Tahoe, CA.

....contains nine words a, href, http, golgi, harvard, edu, biopages, all, and html. All words are converted to upper case. Not all words that appear in a HTML document are used as features. We use an information based approach, similar to that used by an early version of the NewsWeeder program (Lang, 1995) to determine which words to use as features. Intuitively, one would like words that occur frequently in pages on the hotlist, but infrequently on pages on the coldlist (or vice versa) This requires finding E, the expected information content of a word: where p(W) is the probability that a word ....

Lang, K. (1995). NewsWeeder: Learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning. Lake Tahoe, CA.

....algorithms for this task. Let us now discuss the topology of the neural network. We base our representation of the web page on the bag ofwords representation [Salton, 1991] this representation is compatible with neural networks and empirically has proven quite successful in information retrieval [Lang, 1995; Pazzani et al. 1996; Salton, 1991] The bag of words representation simply encodes the frequency of keywords on a page of HTML and ignores word order. We explain below how we obtain a set of key words and phrases) This frequency array serves as the input to the network. Often these ....

K. Lang. NewsWeeder: Learning to Filter News, ICML-95, pp. 331-339.

....Recommender systems have emerged as an important application area and have been the focus of considerable recent academic and commercial interest. The 1997 special issue of the Communications of the ACM [14] contains some key papers. Other important contributions include [2] 4] 8] 13] 16] [9], 1] 12] and [15] In addition, many online retailers are using this technology to recommend new items to their customers, based on what they have bought in the past. Currently, most recommender systems are either content based or collaborative, depending on the type of information that the ....

Ken Lang. Newsweeder: Learning to filter news. In Proceedings of the Twelfth International Conference on Machine Learning, 1997.

....landscape. Information overload is no longer just a popular buzzword, but a daily reality for most of us. This leads to a clear demand for automated methods, commonly referred to as intelligent information agents, that locate and retrieve information with respect to users individual preferences (Lang, 1995; Pazzani and Billsus, 1997; Balabanovic, 1998) As intelligent information agents aim to automatically adapt to individual users, the development of appropriate user modeling techniques is of central importance. Algorithms for intelligent information agents typically draw on work from the ....

Lang, K. (1995). NewsWeeder: learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning. Lake Tahoe, CA, 331--339.

No context found.

K. Lang. Newsweeder: Learning to filter news. In ICML 95, pages 331--339, 1995.

No context found.

Lang, K. (1995). NewsWeeder: Learning to filter news. Intl. Conf. on Machine Learning. p.331-339.

No context found.

K. Lang. Newsweeder: Learning to filter news. In Proceedings of the Twelfth International Conference on Machine Learning, 1995.

No context found.

Ken Lang. 1995. Newsweeder: Learning to filter news. In Proceedings of the Twelfth International Conference on Machine Learning.

No context found.

K. Lang. Newsweeder: Learning to filter news. In ICML 95, pages 331--339, 1995.

No context found.

Lang, K (1995). NewsWeeder: Learning to Filter News. Proceedings of the 12th International Conference on Machine Learning, 331-339, Lake Tahoe, CA.

No context found.

K. Lang. NewsWeeder: Learning to filter news. In Proceedings of the Twelfth International Conference on Machine Learning, pages 331--339. Lake Tahoe, CA, 1995.

No context found.

Lang, K. NEWSWEEDER: Learning to filter news. Proceeding on the 12 th International Conference on Machine Learning, Lake Tahoe, CA, Morgan Kaufmann, pp. 331-339, 1995.

No context found.

Lang, K. (1995). NewsWeeder: Learning to filter news. Proceedings of the Twelfth International Conference on Machine Learning (pp. 331--339). Lake Tahoe, CA.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC