一种两层加权融合的排序算法

当前排序学习算法在学习时将样本集中的所有查询及其相关文档等同对待,忽略了查询之间以及其相关文档之间的差异性,影响了排序模型的性能。对查询之间的差异进行分析,同时考虑文档排序位置造成的资料被检视概率不同的差异特性,提出了一种两层加权融合的排序方法。该方法为每一个查询及其相关文档建立一个子排序模型,在此过程中,对文档赋予非对称权重,然后通过建立新的损失函数作为优化目标,利用损失函数调节不同查询产生损失之间的权重,最终实现多查询相关排序模型的加权融合。在标准数据集LETOR OHSUMED上的实验结果表明,所提方法在排序性能上有较大提升。     

Lightweight integration of IR and DB for scalable hybrid search with integrated ranking support

The Web contains a large amount of documents and an increasing quantity of structured data in the form of RDF triples. Many of these triples are annotations associated with documents. While structured queries constitute the principal means to retrieve structured data, keyword queries are typically used for document retrieval. Clearly, a form of hybrid search that seamlessly integrates these formalisms to query both textual and structured data can address more complex information needs. However, hybrid search on the large scale Web environment faces several challenges. First, there is a need for repositories that can store and index a large amount of semantic data as well as textual data in documents, and manage them in an integrated way. Second, methods for hybrid query answering are needed to exploit the data from such an integrated repository. These methods should be fast and scalable, and in particular, they shall support flexible ranking schemes to return not all but only the most relevant results. In this paper, we present CE², an integrated solution that leverages mature information retrieval and database technologies to support large scale hybrid search. For scalable and integrated management of data, CE² integrates off-the-shelf database solutions with inverted indexes. Efficient hybrid query processing is supported through novel data structures and algorithms which allow advanced ranking schemes to be tightly integrated. Furthermore, a concrete ranking scheme is proposed to take features from both textual and structured data into account. Experiments conducted on DBpedia and Wikipedia show that CE² can provide good performance in terms of both effectiveness and efficiency.     

多查询相关的排序支持向量机融合算法

排序学习是目前信息检索与机器学习领域研究的热点问题.现有排序学习算法在学习时把训练样本集中的所有查询及其相关文档等同对待,忽视了查询之间的差异,影响了排序模型的性能.对查询之间的差异进行描述,并在训练过程中考虑这种差异,提出一种基于有监督学习的融合多个与查询相关排序子模型的方法.该方法为每一个查询及其相关文档建立一个子排序模型,并将子排序模型的输出进行向量化表示,将多个查询相关的排序模型转化为体现查询差异的特征数据,实现多排序模型的集成.以排序支持向量机为例,在查询级和样本级建立新的损失函数作为优化目标,并利用此损失函数调节不同查询产生损失之间的权重,提出多查询相关的排序支持向量机融合算法.在文档检索和网页检索中的实验结果表明,使用多查询相关的排序支持向量机融合算法可以取得比传统排序学习模型更好的性能.     

基于对抗式数据增强的深度文本检索重排序

在信息检索领域的排序任务中, 神经网络排序模型已经得到广泛使用. 神经网络排序模型对于数据的质量要求极高, 但是, 信息检索数据集通常含有较多噪音, 不能精确得到与查询不相关的文档. 为了训练一个高性能的神经网络排序模型, 获得高质量的负样本, 则至关重要. 借鉴现有方法doc2query的思想, 本文提出了深度、端到...     

SEARCHING FOR EXPLANATORY WEB PAGES USING AUTOMATIC QUERY EXPANSION

When one tries to use the Web as a dictionary or encyclopedia, entering some single term into a search engine, the highly ranked pages in the result can include irrelevant or useless sites. The problem is that single-term queries, if taken literally, underspecify the type of page the user wants. For such problems automatic query expansion, also known as pseudo-feedback, is often effective. In this method the top n documents returned by an initial retrieval are used to provide terms for a second retrieval. This paper contributes, first, new normalization techniques for query expansion, and second, a new way of computing the similarity between an expanded query and a document, the "local relevance density" metric, which complements the standard vector product metric. Both of these techniques are shown to be useful for single-term queries, in Japanese, in experiments done over the World Wide Web in early 2001.     

A machine learning approach to query generation in plagiarism source retrieval

Plagiarism source retrieval is the core task of plagiarism detection. It has become the standard for plagiarism detection to use the queries extracted from suspicious documents to retrieve the plagiarism sources. Generating queries from a suspicious document is one of the most important steps in plagiarism source retrieval. Heuristic-based query generation methods are widely used in the current research. Each heuristic-based method has its own advantages, and no one statistically outperforms the others on all suspicious document segments when generating queries for source retrieval. Further improvements on heuristic methods for source retrieval rely mainly on the experience of experts. This leads to difficulties in putting forward new heuristic methods that can overcome the shortcomings of the existing ones. This paper paves the way for a new statistical machine learning approach to select the best queries from the candidates. The statistical machine learning approach to query generation for source retrieval is formulated as a ranking framework. Specifically, it aims to achieve the optimal source retrieval performance for each suspicious document segment. The proposed method exploits learning to rank to generate queries from the candidates. To our knowledge, our work is the first research to apply machine learning methods to resolve the problem of query generation for source retrieval. To solve the essential problem of an absence of training data for learning to rank, the building of training samples for source retrieval is also conducted. We rigorously evaluate various aspects of the proposed method on the publicly available PAN source retrieval corpus. With respect to the established baselines, the experimental results show that applying our proposed query generation method based on machine learning yields statistically significant improvements over baselines in source retrieval effectiveness.     

Comparing data summaries for processing live queries over Linked Data

A growing amount of Linked Data??graph-structured data accessible at sources distributed across the Web??enables advanced data integration and decision-making applications. Typical systems operating on Linked Data collect (crawl) and pre-process (index) large amounts of data, and evaluate queries against a centralised repository. Given that crawling and indexing are time-consuming operations, the data in the centralised index may be out of date at query execution time. An ideal query answering system for querying Linked Data live should return current answers in a reasonable amount of time, even on corpora as large as the Web. In such a live query system source selection??determining which sources contribute answers to a query??is a crucial step. In this article we propose to use lightweight data summaries for determining relevant sources during query evaluation. We compare several data structures and hash functions with respect to their suitability for building such summaries, stressing benefits for queries that contain joins and require ranking of results and sources. We elaborate on join variants, join ordering and ranking. We analyse the different approaches theoretically and provide results of an extensive experimental evaluation.     

文本搜索排序中构造训练集的一种方法

在文本搜索领域,用自学习排序的方法构建排序模型越来越普遍。排序模型的性能很大程度上依赖训练集。每个训练样本需要人工标注文档与给定查询的相关程度。对于文本搜索而言,查询几乎是无穷的,而人工标注耗时费力,所以选择部分有信息量的查询来标注很有意义。提出一种同时考虑查询的难度、密度和多样性的贪心算法从海量的查询中选择有信息量的查询进行标注。在LETOR和从Web搜索引擎数据库上的实验结果,证明利用本文提出的方法能构造一个规模较小且有效的训练集。     

Learning to crawl deep web

Deep web or hidden web refers to the hidden part of the Web (usually residing in structured databases) that remains unavailable for standard Web crawlers. Obtaining content of the deep web is challenging and has been acknowledged as a significant gap in the coverage of search engines. The paper proposes a novel deep web crawling framework based on reinforcement learning, in which the crawler is regarded as an agent and deep web database as the environment. The agent perceives its current state and selects an action (query) to submit to the environment (the deep web database) according to Q-value. While the existing methods rely on an assumption that all deep web databases possess full-text search interfaces and solely utilize the statistics (TF or DF) of acquired data records to generate the next query, the reinforcement learning framework not only enables crawlers to learn a promising crawling strategy from its own experience, but also allows for utilizing diverse features of query keywords. Experimental results show that the method outperforms the state of art methods in terms of crawling capability and relaxes the assumption of full-text search implied by existing methods.     

面向微博搜索的时间敏感的排序学习方法

近年来微博检索已经成为信息检索领域的研究热点。相关的研究表明,微博检索具有时间敏感性。已有工作根据不同的时间敏感性假设,例如,时间越新文档越相关,或者时间越接近热点时刻文档越相关,得到多种不同的检索模型,都在一定程度上提高了检索效果。但是这些假设主要来自于观察,是一种直观简化的假设,仅能从某个方面反映时间因素影响微博排序的规律。该文验证了微博检索具有复杂的时间敏感特性,直观的简化假设并不能准确地描述这种特性。在此基础上提出了一个利用微博的时间特征和文本特征,通过机器学习的方式来构建一个针对时间敏感的微博检索的排序学习模型(TLTR)。在时间特征上,考察了查询相关的全局时间特征以及查询-文档对的局部时间特征。在TREC Microblog Track 20112012数据集上的实验结果表明,TLTR模型优于现有的其他时间敏感的微博排序方法。     

搜索引擎中的聚类浏览技术

搜索引擎大多以文档列表的形式将搜索结果显示给用户,随着Web文档数量的剧增,使得用户查找相关信息变得越来越困难,一种解决方法是对搜索结果进行聚类提高其可浏览性。搜索引擎的聚类浏览技术能使用户在更高的主题层次上查看搜索结果,方便地找到感兴趣的信息。本文介绍了搜索引擎的聚类浏览技术对聚类算法的基本要求及其分类方法,研究分析了主要聚类算法及其改进方法的特点,讨论了对聚类质量的评价,最后指出了聚类浏览技术的发展趋势。     

An evolutionary approach for combining different sources of evidence in search engines

Modern Web search engines use different strategies to improve the overall quality of their document rankings. Usually the strategy adopted involves the combination of multiple sources of relevance into a single ranking. This work proposes the use of evolutionary techniques to derive good evidence combination functions using three different sources of evidence of relevance: the textual content of documents, the reputation of documents extracted from the connectivity information available in the processed collection and the anchor text concatenation. The combination functions discovered by our evolutionary strategies were tested using a collection containing 368 queries extracted from a real nation-wide search engine query log with over 12 million documents. The experiments performed indicate that our proposal is an effective and practical alternative for combining sources of evidence into a single ranking. We also show that different types of queries submitted to a search engine can require different combination functions and that our proposal is useful for coping with such differences.     

A New Retrieval Model Based on TextTiling for Document Similarity Search

Document similarity search is to find documents similar to a given query document and return a ranked list of similar documents to users, which is widely used in many text and web systems, such as digital library, search engine, etc. Traditional retrieval models, including the Okapi's BM25 model and the Smart's vector space model with length normalization, could handle this problem to some extent by taking the query document as a long query. In practice, the Cosine measure is considered as the best model for document similarity search because of its good ability to measure similarity between two documents. In this paper, the quantitative performances of the above models are compared using experiments. Because the Cosine measure is not able to reflect the structural similarity between documents, a new retrieval model based on TextTiling is proposed in the paper. The proposed model takes into account the subtopic structures of documents. It first splits the documents into text segments with TextTiling and calculates the similarities for different pairs of text segments in the documents. Lastly the overall similarity between the documents is returned by combining the similarities of different pairs of text segments with optimal matching method. Experiments are performed and results show: 1) the popular retrieval models (the Okapi's BM25 model and the Smart's vector space model with length normalization) do not perform well for document similarity search; 2) the proposed model based on TextTiling is effective and outperforms other models, including the Cosine measure; 3) the methods for the three components in the proposed model are validated to be appropriately employed.     

Building Minority Language Corpora by Learning to Generate Web Search Queries

The Web is a source of valuable information, but the process of collecting, organizing, and effectively utilizing the resources it contains is difficult. We describe CorpusBuilder, an approach for automatically generating Web search queries for collecting documents matching a minority concept. The concept used for this paper is that of text documents belonging to a minority natural language on the Web. Individual documents are automatically labeled as relevant or nonrelevant using a language filter, and the feedback is used to learn what query lengths and inclusion/exclusion term-selection methods are helpful for finding previously unseen documents in the target language. Our system learns to select good query terms using a variety of term scoring methods. Using odds ratio scores calculated over the documents acquired was one of the most consistently accurate query-generation methods. To reduce the number of estimated parameters, we parameterize the query length using a Gamma distribution and present empirical results with learning methods that vary the time horizon used when learning from the results of past queries. We find that our system performs well whether we initialize it with a whole document or with a handful of words elicited from a user. Experiments applying the same approach to multiple languages are also presented showing that our approach generalizes well across several languages regardless of the initial conditions.     

Using Semantic Context to Rank the Results of Keyword Search

In an empirical user study, we assessed two approaches to ranking the results from a keyword search using semantic contextual match based on Latent Semantic Analysis. These techniques involved searches initiated from words found in a seed document within a corpus. The first approach used the sentence around the search query in the document as context while the second used the entire document. With a corpus of 20,000 documents and a small proportion of relevant documents (<0.1%), both techniques outperformed a conventional keyword search on a recall-based information retrieval (IR) task. These context-based techniques were associated with a reduction in the number of searches conducted, an increase in users’ precision and, to a lesser extent, an increase in recall. This improvement was strongest when the ranking was based on document, rather than sentence, context. Individuals were more effective on the IR task when the lists returned by the techniques were ranked better. User performance on the task also correlated with achievement on a generalized IQ test but not on a linguistic ability test.     

A unified probabilistic framework for Web page scoring systems

The definition of efficient page ranking algorithms is becoming an important issue in the design of the query interface of Web search engines. Information flooding is a common experience especially when broad topic queries are issued. Queries containing only one or two keywords usually match a huge number of documents, while users can only afford to visit the first positions of the returned list, which do not necessarily refer to the most appropriate answers. Some successful approaches to page ranking in a hyperlinked environment, like the Web, are based on link analysis. We propose a general probabilistic framework for Web page scoring systems (WPSS), which incorporates and extends many of the relevant models proposed in the literature. In particular, we introduce scoring systems for both generic (horizontal) and focused (vertical) search engines. Whereas horizontal scoring algorithms are only based on the topology of the Web graph, vertical ranking also takes the page contents into account and are the base for focused and user adapted search interfaces. Experimental results are reported to show the properties of some of the proposed scoring systems with special emphasis on vertical search.     

NERank+: a graph-based approach for entity ranking in document collections

Most entity ranking research aims to retrieve a ranked list of entities from a Web corpus given a user query. The rank order of entities is determined by the relevance between the query and contexts of entities. However, entities can be ranked directly based on their relative importance in a document collection, independent of any queries. In this paper, we introduce an entity ranking algorithm named NERank+. Given a document collection, NERank+ first constructs a graph model called Topical Tripartite Graph, consisting of document, topic and entity nodes. We design separate ranking functions to compute the prior ranks of entities and topics, respectively. A meta-path constrained random walk algorithm is proposed to propagate prior entity and topic ranks based on the graph model.We evaluate NERank+ over real-life datasets and compare it with baselines. Experimental results illustrate the effectiveness of our approach.     

基于分层抽样的重叠深网数据源选择

深网查询在Web上众多的应用,需要查询大量的数据源才能获得足够的数据,如多媒体数据搜索、团购网站信息聚合等.应用的成功,取决于查询多数据源的效率和效果.当前研究侧重查询与数据源的相关性而忽略数据源之间的重叠关系,使得不同数据源上相同结果的数据被重复查询,增加了查询开销及数据源的工作负载.为了提高深网查询的效率,提出一种元组水平的分层抽样方法来估计和利用查询在数据源上的统计数据,选择高相关、低重叠的数据源.该方法分为两个阶段：离线阶段,基于元组水平对数据源进行分层抽样,获得样本数据;在线阶段,基于样本数据迭代地估计查询在数据源上的覆盖率和重叠率,并采用一种启发式策略以高效地发现低重叠的数据源.实验结果表明,该方法能够显著提高重叠数据源选择的精度和效率.     

Beyond Single-Page Web Search Results

Given a user keyword query, current Web search engines return a list of individual Web pages ranked by their "goodness" with respect to the query. Thus, the basic unit for search and retrieval is an individual page, even though information on a topic is often spread across multiple pages. This degrades the quality of search results, especially for long or uncorrelated (multitopic) queries (in which individual keywords rarely occur together in the same document), where a single page is unlikely to satisfy the user's information need. We propose a technique that, given a keyword query, on the fly generates new pages, called composed pages, which contain all query keywords. The composed pages are generated by extracting and stitching together relevant pieces from hyperlinked Web pages and retaining links to the original Web pages. To rank the composed pages, we consider both the hyperlink structure of the original pages and the associations between the keywords within each page. Furthermore, we present and experimentally evaluate heuristic algorithms to efficiently generate the top composed pages. The quality of our method is compared to current approaches by using user surveys. Finally, we also show how our techniques can be used to perform query-specific summarization of Web pages.