Computing semantic relatedness of natural language texts requires access to vast amounts of common-sense and domain-specific world knowledge. We propose Explicit Semantic Analysis (ESA), a novel method that represents the meaning of texts in a high-dimensional space of concepts derived from Wikipedia. We use machine learning techniques to explicitly represent the meaning of any text as a weighted vector of Wikipedia-based concepts. Assessing the relatedness of texts in this space amounts to comparing the corresponding vectors using conventional metrics (e.g., cosine). Compared with the previous state of the art, using ESA results in substantial improvements in correlation of computed relatedness scores with human judgments: from r =0.56 to 0.75 for individual words and from r =0.60 to 0.72 for texts. Importantly, due to the use of natural concepts, the ESA model is easy to explain to human users. 1

An important class of searches on the world-wide-web has the goal to find an entry page (homepage) of an organisation. Entry page search is quite different from Ad Hoc search. Indeed a plain Ad Hoc system performs disappointingly. We explored three non-content features of web pages: page length, number of incoming links and URL form. Especially the URL form proved to be a good predictor. Using URL form priors we found over 70% of all entry pages at rank 1, and up to 89% in the top 10. Non-content features can easily be embedded in a language model framework as a prior probability.

The construction of a text classifier usually involves (i) a phase of term selection, in which the most relevant terms for the classification task are identified, (ii) a phase of term weighting, in which document weights for the selected terms are computed, and (iii) a phase of classifier learning, in which a classifier is generated from the weighted representations of the training documents. This process involves an activity of supervised learning, in which information on the membership of training documents in categories is used. Traditionally, supervised learning enters only phases (i) and (iii). In this paper we propose instead that learning from training data should also affect phase (ii), i.e. that information on the membership of training documents to categories be used to determine term weights. We call this idea supervised term weighting (STW). As an example, we propose a number of "supervised variants" of tfidf weighting, obtained by replacing the idf function with the function that has been used in phase (i) for term selection. We present experimental results obtained on the standard Reuters-21578 benchmark with one classifier learning method (support vector machines), three term selection functions (information gain, chi-square, and gain ratio), and both local and global term selection and weighting.

Empirical studies of information retrieval methods show that good retrieval performance is closely related to the use of various retrieval heuristics, such as TF-IDF weighting. One basic research question is thus what exactly are these "necessary" heuristics that seem to cause good retrieval performance. In this paper, we present a formal study of retrieval heuristics. We formally define a set of basic desirable constraints that any reasonable retrieval function should satisfy, and check these constraints on a variety of representative retrieval functions. We find that none of these retrieval functions satisfies all the constraints unconditionally. Empirical results show that when a constraint is not satisfied, it often indicates non-optimality of the method, and when a constraint is satisfied only for a certain range of parameter values, its performance tends to be poor when the parameter is out of the range. In general, we find that the empirical performance of a retrieval formula is tightly related to how well it satisfies these constraints. Thus the proposed constraints provide a good explanation of many empirical observations and make it possible to evaluate any existing or new retrieval formula analytically.

Text retrieval systems store agreat variety of documents, from abstracts, newspaper articles, and Web pages to journal articles, books, court transcripts, and legislation. Collections of diverse types of documents expose shortcomings in current approaches to ranking. Use of short fragments of documents, called passages, instead of whole documents can overcome these shortcomings: passage ranking provides convenient units of text to return to the user, can avoid the difficulties of comparing documents of different length, and enables identificationofshortblocksofrelevantmaterialamong otherwise irrelevant text. In this article, we compare severalkindsofpassageinanextensiveseriesofexperiments. We introduce anew type of passage, overlapping fragments of either fixed or variable length. We show that ranking with these arbitrary passages gives substantial improvements in retrieval effectiveness over traditional document ranking schemes, particularly for queries on collections of long documents. Ranking with arbitrary passages shows consistent improvements compared to ranking with whole documents, and to ranking with previous passage types that depend on document structure or topic shifts in documents.

Queries to text collections are resolved by ranking the documents in the collection and returning the highest-scoring documents to the user. An alternative retrieval method is to rank passages, that is, short fragments of documents, a strategy that can improve effectiveness and identify relevant material in documents that are too large for users to consider as a whole. However, ranking of passages can considerably increase retrieval costs. In this paper we explore alternative query evaluation techniques, and develop new techniques for evaluating queries on passages. We show experimentally that, appropriately implemented, effective passage retrieval is practical in limited memory on a desktop machine. Compared to passage ranking with adaptations of current document ranking algorithms, our new "DO-TOS" passage ranking algorithm requires only a fraction of the resources, at the cost of a small loss of effectiveness.

Do improvements in system performance demonstrated by batch evaluations conJbr the same benefit for real users? We carried out experiments designed to investigate this question. After identi~ing a weighting scheme that gave maximum improvement over the baseline in a non-interactive evaluation, we used it with real users searching on an instance recall task. Our results showed the weighting scheme giving beneficial results in batch studies did not do so with real users. Further analysis did identi~ other factors predictive of instance recall, including number of documents saved by the user, document recall, and number of documents seen by the user. 1.

The business of Web search, a $10 billion industry, relies heavily on sponsored search, whereas a few carefully-selected paid advertisements are displayed alongside algorithmic search results. A key technical challenge in sponsored search is to select ads that are relevant for the user’s query. Identifying relevant ads is challenging because queries are usually very short, and because users, consciously or not, choose terms intended to lead to optimal Web search results and not to optimal ads. Furthermore, the ads themselves are short and usually formulated to capture the reader’s attention rather than to facilitate query matching. Traditionally, matching of ads to queries employed standard information retrieval techniques using the bag of words approach. Here we propose to go beyond the bag of words, and augment both queries and ads with additional knowledgerich features. We use Web search results initially returned for the query to create a pool of relevant documents. Classifying these documents with respect to an external taxonomy and identifying salient named entities give rise to two new feature types. Empirical evaluation based on over 9,000 query-ad pairwise judgments confirms that using augmented queries produces highly relevant ads. Our methodology also relaxes the requirement for each ad to explicitly specify the exhaustive list of queries (“bid phrases”) that can trigger it.

Text categorization (also known as text classification, or topic spotting) is the task of automatically sorting a set of documents into categories from a predefined set. This task has several applications, including automated indexing of scientific articles according to predefined thesauri of technical terms, filing patents into patent directories, selective dissemination of information to information consumers, automated population of hierarchical catalogues of Web resources, spam filtering, identification of document genre, authorship attribution, survey coding, and even automated essay grading. Automated text classification is attractive because it frees organizations from the need of manually organizing document bases, which can be too expensive, or simply not feasible given the time constraints of the application or the number of documents involved. The accuracy of modern text classification systems rivals that of trained human professionals, thanks to a combination of information retrieval (IR) technology and machine learning (ML) technology. This chapter will outline the fundamental traits of the technologies involved, of the applications that can feasibly be tackled through text classification, and of the tools and resources that are available to the researcher and developer wishing to take up these technologies for deploying real-world applications. 1

Malicious programs, such as viruses and worms, are frequently related to previous programs through evolutionary relationships. Discovering those relationships and constructing a phylogeny model is expected to be helpful for analyzing new malware and for establishing a principled naming scheme. Matching permutations of code may help build better models in cases where malware evolution does not keep things in the same order. We describe method for constructing phylogeny models that uses features called n-perms to match possibly permuted code. An experiment was performed to compare the relative effectiveness of vector similarity measures using n-perms and n-grams when comparing permuted variants of programs. The similarity measures using n-perms maintained a greater separation between the similarity scores of permuted families of specimens versus unrelated specimens. A subsequent study using a tree generated through suggests that phylogeny models based on may help forensic analysts investigate new specimens, and assist in reconciling malware naming inconsistencies.