Multi-Instance Learning Based Web Mining

Abstract

In multi-instance learning, the training set comprises labeled bags that are composed of unlabeled instances, and the task is to predict the labels of unseen bags. In this paper, a web mining problem, i.e. web index recommendation, is investigated from a multiinstance view. In detail, each web index page is regarded as a bag, while each of its linked pages is regarded as an instance. A user favoring an index page means that he or she is interested in at least one page linked by the index. Based on the browsing history of the user, recommendation could be provided for unseen index pages. An algorithm named Fretcit-kNN, which employs the Minimal Hausdorff distance between frequent term sets and utilizes both the references and citers of an unseen bag in determining its label, is proposed to solve the problem. Experiments show that in average the recommendation accuracy of Fretcit-kNN is 81.0% with 71.7% recall and 70.9% precision, which is significantly better than the best algorithm that does not consider the specific characteristics of multi-instance learning, whose performance is 76.3% accuracy with 63.4% recall and 66.1% precision.

On the relation between multi-instance learning and semi-supervised learning

“…Multi-instance learning techniques have already been applied to diverse applications such as image categorization (Maron & Ratan, 1998;Chen & Wang, 2004;Chen et al, 2006), computer security (Ruffo, 2000), Web mining (Zhou et al, 2005b), face detection (Viola et al, 2006), etc.…”

“…The Web index page recommendation data sets described in (Zhou et al, 2005b) were used in this experiment. A Web index page is a Web page which contains plentiful information but only provides titles or brief summaries while leaving the detailed presentation to its linked pages.…”

“…For each of the nine data sets, 75 bags were randomly selected for training while the remaining 38 bags were used for testing. We used the processed data 2 such that all the bags and instances are as same as those used in (Zhou et al, 2005b). The number of positive and negative bags in the data sets is tabulated in Table 4.…”

“…The performance of MissSVM measured by precision, recall and F-measure on these data sets are depicted in Figure. 2. The figure also includes the performance of Fretcit-kNN, r-Fretcit-kNN and TFIDF reported in literature (Zhou et al, 2005b). The results averaged across all the nine data sets are summarized in Table 5.…”

Minimizing Size of Decision Trees for Multi-label Decision Tables

“…N OW a days, multi-label decision tables have gained attention in problem of semantic annotation of images [1], music categorization into emotions [2], functional genomics [3], and text categorization [4]. Furthermore, it is natural to have such data sets in optimization problems such as finding a Hamiltonian circuit with the minimum length in the traveling salesman problem [5], finding nearest post office in the post office problem [5]; in this case we give input with more than one optimal solutions.…”