Balabanovic, M.: An interface for learning multi-topic user profiles from implicit feedback. Proc. of AAAI Workshop on Recommender Systems. Madison, WI, USA (1998)  Home/Search   Document Details and Download   Summary   Related Articles   Check

....is defined as the frequency (TfIdf and similar) of those phrases in the documents the user has rated ( relevance feedback ) as interesting in the past. In order to decide whether to recommend a document or not, the document (i.e. the corresponding vector) is compared to the user model vector, see [2, 1, 9], 5] and [6] Though the vectors represent a user s interest, they do not explicitely describe a user s interest: A sequence of word frequencies is not a user model that can be explained to the user in an intutive way. Thus, our motivation was to find a transparent formalism which is accessible ....

....by concepts Since a user is not interested in single, very specific topics only, a user model may consist of several aspects which describe di#erent, specific parts of a user s interest. Aspects can be compared to specific topics of interest, or, e.g. to folders for news items or e mails (c.f. [1, 12]) The aspect shown in figure 2 is interpreted as: A document d is considered relevant with respect to the aspect Conceptual User Models if it classifies as: 1. a publication about knowledge representation 2. a publication about symbolic machine learning 3. a publication about user ....

Balabanovic, M. An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems (Madison, Wisconsin, July 1998).

....the system. While this provides a high level of detail about the user, explicit modelling is often seen as time consuming and intrusive [13, 15] The alternative technique, implicit modelling, obtains user preferences through observations of the users activities as they interact with the system [2]. However implicit modelling provides only positive exemplars of user actions and requires both lengthy user interaction and the ability to track users over multiple sessions [18] The Hyperdoc system does not currently provide enough interaction to allow implicit user modelling. The compromise ....

M. Balabanovic. An Interface for Learning Multi-topic User Profiles from Implicit Feedback. In Proceedings of the AAAI-98 Workshop on Recommender Systems, Madison, Wisconsin, 1998.

.... ratings have failed so far [KRBH98] Ratings may be provided explicitly, e.g. by selecting a rating from a pull down menu or by pressing a numeric key, or implicitly by monitoring user activities, e.g. reading time of news messages [MS94] or the saving of an object for later use [Nic97, AZ97, Bal98] Similar information may also be gathered in an offline manner by mining objects containing usage data. Examples for such systems are the Referral Web [KSS97a, KSS97b, KS98] that mines publicly available documents to build a model of existing real world relations between users, and the Phoaks ....

M. Balabanovic. An interface for learning multi-topic user profiles from implicit feedback. In Recommender Systems, Papers from the 1998 Workshop,

.... out about the Benedictine order would use the document in a different context from a person interested in cooking, and in each context, there would be a different answer to the question What is this document about Information retrieval agents generally retrieve based on document content [3, 6, 7, 14, 16], and commonly treat a document as an independent entity, taking into account its relationship to the entire corpus but not to the immediate group of recently consulted 1 For a contrasting view, see [8] documents within which it was accessed. For example, in TFIDF based indexing, an index ....

Marko Balabanovic. An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems, 1998.

....filtering systems that rely on explicit feedback # usually corresponds to the rating a user has given to a document and can either be a positive or a negative value. Systems that use implicit feedback may use different values for # for different kinds of feedback. In the recommender system Slider [Balabanovic 1998] for example, # is set to 3 when a user has deleted an article and to 0.5 when a users has read an article. Because PRES only determines whether or not a document is relevant, weight # is always set to 1. The profile vector is adjusted to a diminishing of the user s interests by #, a weight ....

Balabanovic, M. (1998). An Interface for Learning Multi-topic User Profiles from Implicit Feedback. AAAI-98 Workshop on Recommender Systems, Madison, Wisconsin.

....judgments on retrieved documents. Among other reasons, users may have problems to decide about relevance of some documents. An alternative is to use implicit feedback, i.e. to infer document relevance from user s behavior, which has been successfully applied in the learning of user models [4]. Then, we use the documents read by the user as feedback. The system provides numerous context elements, including a user adapted summary that can assist users to decide about document relevance without inspecting the full text. If a user accesses the full text of a piece of news, the system can ....

Balabanovic, M.: An Interface for Learning Multi-topic User Profiles from Implicit Feedback. In: AAAI Workshop on Recommender Systems, Madison, Wisconsin (1998)

....and ontology refinement processes within OySTER. Further technicalities had to be truncated with respect to limited space. 1 Adaptive Information Retrieval from the Web Efficient User Modeling in Recommender Systems. In course of building recommender systems for the Www, many projects (c.f. [1, 2, 6, 12]) tackled the problem of personalized document filtering by using n ary vectors representing the user s interest. This implies an intertwined representation of information resources (namely the representation of a web document as a vector) and user models. Though vectors represent a user s ....

....Only few users will be provide at least partial feedback about delivered search results. Thus we use different techniques in order to receive labeled training data: On the one hand, we will use a Slider like interface in order to receive hidden explicit feedback on search results (c.f. [1]) But since this will provide only a small amount of feedback, we will have to make use of more sophisticated methods for generating labelled data. 4 Note, that this case could be a follow up of the former case, if u and v actually turn out to have different interests. This can easliy be ....

Balabanovic, M. An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems (Madison, Wisconsin, July 1998).

....on Ml4Um ) Wrong algorithms that produce bad results, perform best . interesting in the past. Vector length is reduced by extracting most relevant phrases using TfIDf measures and labeling of examples is achieved by relevance feedback which assigns target function values to vectors, see [BP99, Bal98, BP97, JFM97, PMB96, Lie95] In order to decide whether to recommend a document or not, the current document s classification (i.e. the corresponding vector) is compared to the user model vector. Our idea is to distinguish between document classification and user model. The core ideas behind the ....

....as indicated in figure 3. 3 Learning Conceptual User Models Knowing and learning about the user allows for a more precise information extraction. This has been shown in various web based user specific agent systems like WebWatcher [JFM97] Syskill Webert [PMB96] Letitzia [Lie95] and Fab [Bal98] In our system we shall investigate an approach for order sorted Ilp (as described in [Mul95] for inducing user models. Thus, we are able to perform content based classification [Kru95] that goes beyond statistical phrase clustering. User model induction. Supervised learning algorithms need ....

[Article contains additional citation context not shown here]

M. Balabanovic. An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems, Madison, Wisconsin, July 1998.

....search engines. This has been shown in various web based user specific agent systems. Most of them are a kind of recommendation system, like WebWatcher [Joachims et al. 1997] Syskill Webert [Pazzani et al. 1996] Letitzia [Lieberman, 1995] and Fab [Balabanovic et al. 1997] The Slider, c. f. [Balabanovic, 1998], employs a very cute method for receiving explicit user feedback by interpreting operations on topic lists as positive and negative feedback. 3 Though it is claimed by nearly any project as a short term goal. Sadly, user models are mostly described by keyword patterns, that is, by ....

....Thus, explicit labeling is hidden under a sorting routine for a search engine generated hitlist: The user can hide, fold and unfold nodes from the result page to generate a bookmarks file where his actions are interpreted as positive or negative feedback for the Urls. c. f. Slider interface, [Balabanovic, 1998]) ffl Finally, instead of merely generating individual user models, we make use of collaborative user modeling, where good and bad examples of similar user profiles are mutually exchanged in order to both refine the user models as well as the document category descriptions. Learning Tasks: User ....

Balabanovic, M. (1998). An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems, Madison, Wisconsin.

.... Thus, explicit labelling is hidden under a sorting routine for a search engine generated hitlist: The user can hide, fold and unfold nodes from the result page to generate a bookmarks file where his actions are interpreted as positive and negative feedback for the Urls (see the Slider interface, [1]) Finally, instead of only generating individual user models, we make use of collaborative user modeling, where good and bad examples of similar user profiles are mutually exchanged in order to both refine the user models and the document category descriptions (see [2] Further information ....

M. Balabanovic. An interface for learning multi-topic user profiles from implicit feedback. In AAAI-98 Workshop on Recommender Systems, Madison, Wisconsin, July 1998.

No context found.

Balabanovic, M.: An interface for learning multi-topic user profiles from implicit feedback. Proc. of AAAI Workshop on Recommender Systems. Madison, WI, USA (1998)

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