P. Bollmann and S.K.M. Wong. Adaptive linear information retrieval models. In Proceedings of the ACM SIGIR Conference on Research and Development in Information Retrieval, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... prefer one to other or regard both as being equivalent with respect to her information needs (or search queries) In other words, user preference of documents in D de nes a preference relation over D as follows: 8d; d 0 2 D; d d 0 ( the user prefers d 0 to d: It has been shown in [1] that if a user preference relation is a weak order satisfying some additional conditions then it can be represented by a linear classi er. That is, there is a query vector q = q 1 ; q n ) 2 R n such that 8d; d 0 2 D; d d 0 ( q d q d 0 : 1) In general a linear classi ....

P. Bollmann and S.K.M. Wong. Adaptive linear information retrieval models. In Proceedings of the ACM SIGIR Conference on Research and Development in Information Retrieval, 1987.

....closer to the user ranking is better than a system that produces a ranking that is further away. To quantify this idea, a performance measure may be derived by using the distance between a user ranking and a system ranking. A possible evaluation measure is R norm as suggested by Bollmann and Wong [2], other measures have also been proposed [10] Let (D, #) be a document space, where D is a finite set of documents and # be a preference relation as defined above. Let # be some ranking of D given by a retrieval system. Then R norm is defined as R norm (#) 1 2 # 1 S S S ....

Bollman, P. and Wong, S. K. M. Adaptive linear information retrieval models. In Proc. of the 10 th Int. ACM SIGIR Conference on Research and Development in Information Retrieval, pages 157-163, New Orleans, LA., June 1987.

....judgments were made independently in uncis1 and uncis2 and appear to have undermined all other efforts to produce comparable results; that is, to produce comparisons attributable primarily to the algorithms. adaptive linear model (run uncis1) Run uncis1 employed the adaptive linear (AL) model (Bollmann Wong, 1987; Wong Yao, 1990; Wong, Yao, Bollmann, 1988; Wong, Yao, Salton, Buckley, 1991) The AL model is based on the concept of the preference relation (Fishburn, 1970; Roberts, 1979) The user preference relation on a set of documents D is defined as a binary relation on D where for all d, d D, ....

.... Wong, Yao, Salton, Buckley, 1991) The AL model is based on the concept of the preference relation (Fishburn, 1970; Roberts, 1979) The user preference relation on a set of documents D is defined as a binary relation on D where for all d, d D, d d the user with a query prefers d to d (5) (Bollmann Wong, 1987). An IR model based on the user preference relation allows the use of a multivalued relevance scale for example, relevant, marginally relevant, and not relevant. However, for TREC 5, only binary relevance distinctions were made. Wong et al. 1991) assume that on D can be mapped to a ....

Bollmann, P., & Wong, S. K. M. (1987). Adaptive linear information retrieval models.

....the users is binary (i.e. relevant or non relevant) and the term weights are numeric. In contrast, in the context of deterministic relevance feedback techniques, algorithms for deriving optimal retrieval strategies are known even when the user feedback is given using multiple levels of relevance [8, 9]. More speci cally, the Perceptron convergence algorithm has been extended to deal with weighted terms and when the user feedback is given as a preference relation de ned over documents, with respect to the given user need. In this paper, we provide a generalization of the term relevance weight, ....

....weight that is used for determining the search term weight of j th term of q along with its term frequency [12] 2. 2 Document Space and Query Space By a document space, we mean a relational system (D; where D is a set of document descriptions and is a preference relation on D [9, 8]. A document description stands for a nonempty equivalence class of documents for all those with the same description. The preference relation is de ned relative to the information need of some user. We assume that, for the preference relation, the following two conditions hold for all d; d ....

P. Bollmann and S. K. M. Wong, \Adaptive linear information retrieval models," in Proceedings of the Tenth International Conference on Research and Development in Information Retrieval, (New Orleans, LA), pp. 157-163, June 1987. 12

....weight that is used for determining the search term weight of j th term of q along with its term frequency [2] 1. 2 Document Space and Query Space By a document space, we mean a relational system (D; Delta ) where D is a set of document descriptions and Delta is a preference relation on D [5, 6]. A document description stands for a nonempty equivalence class of documents for all those with the same description. The preference relation Delta is defined relative to the information need of some user. We assume that, for the preference relation, the following two conditions hold for all ....

....space (D; Delta ) the retrieval function F : D Theta Q R; which maps each documentquery pair into the real numbers, and the evaluation measure that assesses the rankings generated by F: We illustrate this in the following example. In the example, the evaluation measure R norm is chosen [6]. Example 2.2: We consider the document space of the previous example. Let Q = D: The retrieval function F is given by the coefficient of Jaccard, where F (d; q) # of terms in d and q # of terms in d or q : The evaluation measure is the normalized recall, R norm : Let q Gamma (1; 1; 0) ....

P. Bollmann and S. K. M. Wong, "Adaptive linear information retrieval models, " in Proceedings of the Tenth International Conference on Research and Development in Information Retrieval, (New Orleans, LA), pp. 157--163, June 1987.

....this reduces to learning equivalence relations from examples. But there is also the view that the similarity of the documents to the query represents the importance of the documents (Salton 1989, p. 317) which in turn means that a user need implies some preference relation on the documents. In (Bollmann Wong 1987) and (Wong, Yao, Bollmann 1988) the idea was developed to learn a preference relation instead of an equivalence relation. The learning of preference relations reduces to a standard classification problem if pairs of objects are considered, because a binary relation can be viewed as a subset of ....

Bollmann, P., and Wong, S. K. M. 1987. Adaptive linear information retrieval models. In Proceedings of the 10th Annual ACM SIGIR Conference an Research and Development in Information Retrieval, 157--163.

No context found.

P. Bollmann and S. K. M. Wong. Adaptive Linear Information Retrieval Models. Proc. of the Tenth Annual International ACM/SIGIR Conf. on Research and Development in Information Retrieval, New Orleans, LA, June 1987, pp. 157--163.

....the relevant and nonrelevant documents is independent in a probablistic sense and the user wants to minimize expected loss . From these empirical conditions, a retrieval function can be derived that gives optimal retrieval. A similar approach was also used in the case of linear retrieval functions [BoWo87]. Further examples of this approach to IR theory are found in [RoMC82, YuSS78, YuLC76, ChYu82, RaJu89] We see that if the theory is elaborated sufficiently, this results in an IRS model and hence in an IRS. More generally, we note that the benefits of having an elaborated IR theory are twofold. ....

....the reality. The axiomatic approach makes these assumptions explicit. Hence, given some retrieval model, our goal is to derive necessary and sufficient empirical conditions that the system will work properly. These conditions are also called axioms. An example of this approach can be found in [BoWo87], where necessary and sufficient conditions are given in order to have perfect retrieval in some special cases. It is proven, assuming binary document representation and the use of dot product as the retrieval function, that perfect retrieval is possible only if document (ff 1 ; ff ....

P. Bollmann and S.K.M. Wong. Adaptive linear information retrieval models. Tenth International ACM SIGIR Conf., New Orleans, LA, June 1987, 157-163.

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