Entity matching (a.k.a. record linkage) plays a crucial role in integrating multiple data sources, and numerous matching solutions have been developed. However, the solutions have largely exploited only information available in the mentions and employed a single matching technique. We show how to exploit information about data sources to significantly improve matching accuracy. In particular, we observe that different sources often vary substantially in their level of semantic ambiguity, thus requiring different matching techniques. In addition, it is often beneficial to group and match mentions in related sources first, before considering other sources. These observations lead to a large space of matching strategies, analogous to the space of query evaluation plans considered by a relational optimizer. We propose viewing entity matching as a composition of basic steps into a “match execution plan”. We analyze formal properties of the plan space, and show how to find a good match plan. To do so, we employ ideas from social network analysis to infer the ambiguity and relatedness of data sources. We conducted extensive experiments on several real-world data sets on the Web and in the domain of personal information management (PIM). The results show that our solution significantly outperforms current best matching methods. 1.

A large number of web pages contain data structured in the form of “lists”. Many such lists can be further split into multi-column tables, which can then be used in more semantically meaningful tasks. However, harvesting relational tables from such lists can be a challenging task. The lists are manually generated and hence need not have well defined templates – they have inconsistent delimiters (if any) and often have missing information. We propose a novel technique for extracting tables from lists. The technique is domain-independent and operates in a fully unsupervised manner. We first use multiple sources of information to split individual lines into multiple fields, and then compare the splits across multiple lines to identify and fix incorrect splits and bad alignments. In particular, we exploit a corpus of HTML tables, also extracted from the Web, to identify likely fields and good alignments. For each extracted table, we compute an extraction score that reflects our confidence in the table’s quality. We conducted an extensive experimental study using both real web lists and lists derived from tables on the Web. The experiments demonstrate the ability of our technique to extract tables with high accuracy. In addition, we applied our technique on a large sample of about 100,000 lists crawled from the Web. The analysis of the extracted tables have led us to believe that there are likely to be tens of millions of useful and query-able relational tables extractable from lists on the Web. 1.

A search engine returned result page may contain search results that are organized into multiple dynamically generated sections in response to a user query. Furthermore, such a result page often also contains information irrelevant to the query, such as information related to the hosting site of the search engine. In this paper, we present a method to automatically generate wrappers for extracting search result records from all dynamic sections on result pages returned by search engines. This method has the following novel features: (1) it aims to explicitly identify all dynamic sections, including those that are not seen on sample result pages used to generate the wrapper, and (2) it addresses the issue of correctly differentiating sections and records. Experimental results indicate that this method is very promising. Automatic search result record extraction is critical for applications that need to interact with search engines such as automatic construction and maintenance of metasearch engines and deep Web crawling. 1.

Abstract. This paper studies automatic extraction of structured data from Web pages. Each of such pages may contain several groups of structured data records. Existing automatic methods still have several limitations. In this paper, we propose a more effective method for the task. Given a page, our method first builds a tag tree based on visual information. It then performs a post-order traversal of the tree and matches subtrees in the process using a tree edit distance method and visual cues. After the process ends, data records are found and data items in them are aligned and extracted. The method can extract data from both flat and nested data records. Experimental evaluation shows that the method performs the extraction task accurately. 1

Abstract. In this paper, we present a taxonomy-driven approach to the extraction of data records from web pages containing multiple similar items. In our approach, we first automatically extract a taxonomy from the web for the target domain, then extract data records from web pages in the target domain, and finally use the automatically-extracted taxonomy to automatically annotate feature values in the text of each data record. A key advantage of our approach is that it permits data extraction without human labeling or training. Our evaluation results show that our approach performs well for this task. 1

Abstract. We present a novel approach to automatic information extraction from Deep Web Life Science databases using wrapper induction. Traditional wrapper induction techniques focus on learning wrappers based on examples from one class of Web pages, i.e. from Web pages that are all similar in structure and content. Thereby, traditional wrapper induction targets the understanding of Web pages generated from a database using the same generation template as observed in the example set. However, Life Science Web sites typically contain structurally diverse web pages from multiple classes making the problem more challenging. Furthermore, we observed that such Life Science Web sites do not just provide mere data, but they also tend to provide schema information in terms of data labels – giving further cues for solving the Web site wrapping task. Our solution to this novel challenge of Site-Wide wrapper induction consists of a sequence of steps: 1. classification of similar Web pages into classes, 2. discovery of these classes and 3. wrapper induction for each class. Our approach thus allows us to perform unsupervised information retrieval from across an entire Web site. We test our algorithm against three real-world biochemical deep Web sources and report our preliminary results, which are very promising.

A new approach has been developed for acquiring bilingual web pages from the result pages of search engines, which is composed of two challenging tasks. The first task is to detect web records embedded in the result pages automatically via a clustering method of a sample page. Identifying these useful records through the clustering method allows the generation of highly effective features for the next task which is high-quality bilingual web page acquisition. The task of high-quality bilingual web page acquisition is a classification problem. One advantage of our approach is that it is search engine and domain independent. The test is based on 2516 records extracted from six search engines automatically and annotated manually, which gets a high precision of 81.3 % and a recall of 94.93%. The experimental results indicate that our approach is very effective. 1

Abstract not found