Search Snippets Research Articles

A hybrid classification method via keywords screening and attention mechanisms in extreme short text

Short text classification has provoked a vast amount of attention and research in recent decades. However, most existing methods only focus on the short texts that contain dozens of words like Twitter and Microblog, while pay far less attention to the extreme short texts like news headline and search snippets. Meanwhile, contemporary short text classification methods that extend the features via external knowledge sources always introduce lots of useless concepts, which may be detrimental to classification performance. Moreover, unlike traditional short text classification methods, the classification results of extreme short texts are often determined by a few even one or two keywords. To address these problems, we propose a novel hybrid classification method via Keywords Screening and Attention Mechanisms in extreme short text, called KSAM. More specifically, firstly, the attention-based BiLSTM is introduced in our method to enhance the role of keywords. Secondly, we screen the keywords in the extreme short text for obtaining the true class label, and the concepts concerning the keywords are retrieved from external open knowledge sources like DBpedia. Thirdly, the attention mechanisms are introduced to acquire the weight of these retrieved concepts. Finally, conceptual information is utilized to assist the classification of the extreme short text. Extensive experiments have demonstrated the effectiveness of our method compared to other state-of-the-art methods.

Short text topic modelling using local and global word-context semantic correlation.

Nowadays, people use short text to portray their opinions on platforms of social media such as Twitter, Facebook, and YouTube, as well as on e-commerce websites such as Amazon and Flipkart to share their commercial purchasing experiences. Every day, billions of short texts are created worldwide in tweets, tags, keywords, search queries etc. However, this short text possesses inadequate contextual information, which can be ambiguous, sparse, noisy, remains a major challenge. State-of-the-art strategies of topic modeling such as Latent Dirichlet Allocation and Probabilistic Latent Semantic Analysis are not suitable as it contains a limited number of words in a single document. This work proposes a new model named G_SeaNMF (Gensim_SeaNMF) to improve the word-context semantic relationship by using local and global word embedding techniques. Word embeddings learned from a large corpus provide general semantic and syntactic information about words; it can guide topic modeling for short text collections as supporting information for sparse co-occurrence patterns. In the proposed model, SeaNMF (Semantics-assisted Non-negative Matrix Factorization) is incorporated with word2vec model of Gensim library to strengthen the word's semantic relationship. In this article, a short text topic modeling techniques based on DMM (Dirichlet Multinomial Mixture), self-aggregation and global word co-occurrence were explored. These are evaluated using different measures to gauge cluster coherence on real-world datasets such as Search Snippet, Biomedicine, Pascal Flickr, Tweet and TagMyNews. Empirical evaluation shows that a combination of local and global word embedding provides more appropriate words under each topic with improved outcomes.

Understanding credibility judgements for web search snippets

PurposeIt is well known that information behaviour can be biased in countless ways and that users of web search engines have difficulty in assessing the credibility of results. Yet, little is known about how search engine result page (SERP) listings are used to judge credibility and in which if any way such judgements are biased. The paper aims to discuss these issues.Design/methodology/approachTwo studies are presented. The first collects data by means of a controlled, web-based user study (N=105). Studying judgements for three controversial topics, the paper examines the extent to which users agree on credibility, the extent to which judgements relate to those applied by objective assessors and to what extent judgements can be predicted by the users’ position on and prior knowledge of the topic. A second, qualitative study (N=9) utilises the same setup; however, transcribed think-aloud protocols provide an understanding of the cues participants use to estimate credibility.FindingsThe first study reveals that users are very uncertain when assessing credibility and their impressions often diverge from objective judges who have fact checked the sources. Little evidence is found indicating that judgements are biased by prior beliefs or knowledge, but differences are observed in the accuracy of judgements across topics. Qualitatively analysing think-aloud transcripts from participants think-aloud reveals ten categories of cues, which participants used to determine the credibility of results. Despite short listings, participants utilised diverse cues for the same listings. Even when the same cues were identified and utilised, different participants often interpreted these differently. Example transcripts show how participants reach varying conclusions, illustrate common mistakes made and highlight problems with existing SERP listings.Originality/valueThis study offers a novel perspective on how the credibility of SERP listings is interpreted when assessing search results. Especially striking is how the same short snippets provide diverse informational cues and how these cues can be interpreted differently depending on the user and his or her background. This finding is significant in terms of how search engine results should be presented and opens up the new challenge of discovering technological solutions, which allow users to better judge the credibility of information sources on the web.

A New Vector Representation of Short Texts for Classification

Short and sparse characteristics and synonyms and homonyms are main obstacles for short-text classification. In recent years, research on short-text classification has focused on expanding short texts but has barely guaranteed the validity of expanded words. This study proposes a new method to weaken these effects without external knowledge. The proposed method analyses short texts by using the topic model based on Latent Dirichlet Allocation (LDA), represents each short text by using a vector space model and presents a new method to adjust the vector of short texts. In the experiments, two open short-text data sets composed of google news and web search snippets are utilised to evaluate the classification performance and prove the effectiveness of our method.

Beyond Query-Oriented Highlighting: Investigating the Effect of Snippet Text Highlighting in Search User Behavior.

Search users rely on result captions including titles, snippets, and URLs to decide whether they should read and click a particular result or not. Snippet usually serves as a query-dependent summary of its corresponding landing page and is therefore treated as one of the most important factors in search interaction process. Although there exist many efforts in improving snippet generation algorithms and incorporating more powerful interaction functions into snippets, little attention is paid to the effect of text highlighting in user behaviors. The highlighting of query terms in search snippets has been regarded as a matter of course and whether there exists a better way in snippet text highlighting remains uninvestigated. In this paper, we try to find out whether the default strategy of highlighting query terms employed by most commercial search engines is the best for search users. Through carefully designed experiments, we show that the retrieval efficiency can be affected by different term-highlighting strategies without changes in snippet contents. We also propose an automatic method which adopts CRF to learn to highlight terms based on word embedding, Wikipedia, and snippet content information. Experimental results show that the proposed method could predict highlighted terms selected by crowd workers with moderate performance.

Learning From Short Text Streams With Topic Drifts

Short text streams such as search snippets and micro blogs have been popular on the Web with the emergence of social media. Unlike traditional normal text streams, these data present the characteristics of short length, weak signal, high volume, high velocity, topic drift, etc. Short text stream classification is hence a very challenging and significant task. However, this challenge has received little attention from the research community. Therefore, a new feature extension approach is proposed for short text stream classification with the help of a large-scale semantic network obtained from a Web corpus. It is built on an incremental ensemble classification model for efficiency. First, more semantic contexts based on the senses of terms in short texts are introduced to make up of the data sparsity using the open semantic network, in which all terms are disambiguated by their semantics to reduce the noise impact. Second, a concept cluster-based topic drifting detection method is proposed to effectively track hidden topic drifts. Finally, extensive studies demonstrate that as compared to several well-known concept drifting detection methods in data stream, our approach can detect topic drifts effectively, and it enables handling short text streams effectively while maintaining the efficiency as compared to several state-of-the-art short text classification approaches.

Enabling maps/location searches on mobile devices: constructing a POI database via focused crawling and information extraction

ABSTRACTWith the popularity of mobile devices and smartphones, we have witnessed rapid growth in mobile applications and services, especially in location-based services (LBS). According to a mobile marketing survey, maps/location searches are among the most utilized services on smartphones. Points of interest (POIs), such as stores, shops, gas stations, parking lots, and bus stops, are particularly important for maps/location searches. Existing map services such as Google Maps and Wikimapia are constructed manually either professionally or with crowd sourcing. However, manual annotation is costly and limited in current POI search services. With the abundance of information on the Web, many store POIs can be extracted from the Web. In this paper, we focus on automatically constructing a POI database to enable store POI map searches. We propose techniques that are required to construct a POI database, including focused crawling, information extraction, and information retrieval techniques. We first crawl Yellow Page web sites to obtain vocabularies of store names. These vocabularies are then investigated with search engines to obtain sentences containing these store names from search snippets in order to train a store name recognition model. To extract POIs scattered across the Web, we propose a query-based crawler to find address-bearing pages that might be used to extract addresses and store names. We crawled 1.25 million distinct POI pairs scattered across the Web and implemented a POI search service via Apache Lucent’s search platform, called Solr. The experimental results demonstrate that the proposed geographical information retrieval model outperforms Wikimapia and a commercial app called ‘What’s the Number?’

Enhancing POI search on maps via online address extraction and associated information segmentation

With the popularity of wireless networks and mobile devices, we have seen rapid growth in mobile applications and services, especially location-based services. However, most existing location-based services like Google Maps and Wikimapia rely on crowd-sourcing or business-data providers to maintain their points-of-interest (POI) databases, which are slow and insufficient. Because most updated information can be found on the Web, the insufficiency of current POI databases can be complemented by automatically extracting POIs and their descriptions from general webpages. In this study, we enhance location-based search on maps via online address extraction and associated information segmentation. Given a POI query that cannot be found on a map, we propose a method for extracting the address from search snippets of the query to exploit information from the Web. We demonstrate the application of sequence labeling to Chinese postal-address extraction and compare the performance with and without Chinese word segmentation. Meanwhile, we also present a novel algorithm for associated information segmentation by making use of a document-object model (DOM) tree structure based on the farthest distinguishable ancestor (FDA) of each address. The FDA algorithm is able to locate associated information for each Chinese address resulting in an improvement from an F-measure of 0.811 to 0.964.

Source Free Transfer Learning for Text Classification

Transfer learning uses relevant auxiliary data to help the learning task in a target domain where labeled data is usually insufficient to train an accurate model. Given appropriate auxiliary data, researchers have proposed many transfer learning models. How to find such auxiliary data, however, is of little research so far. In this paper, we focus on the problem of auxiliary data retrieval, and propose a transfer learning framework that effectively selects helpful auxiliary data from an open knowledge space (e.g. the World Wide Web). Because there is no need of manually selecting auxiliary data for different target domain tasks, we call our framework Source Free Transfer Learning (SFTL). For each target domain task, SFTL framework iteratively queries for the helpful auxiliary data based on the learned model and then updates the model using the retrieved auxiliary data. We highlight the automatic constructions of queries and the robustness of the SFTL framework. Our experiments on 20NewsGroup dataset and a Google search snippets dataset suggest that the framework is capable of achieving comparable performance to those state-of-the-art methods with dedicated selections of auxiliary data.

A semantics-based method for clustering of Chinese web search results

Information explosion is a critical challenge to the development of modern information systems. In particular, when the application of an information system is over the Internet, the amount of information over the web has been increasing exponentially and rapidly. Search engines, such as Google and Baidu, are essential tools for people to find the information from the Internet. Valuable information, however, is still likely submerged in the ocean of search results from those tools. By clustering the results into different groups based on subjects automatically, a search engine with the clustering feature allows users to select most relevant results quickly. In this paper, we propose an online semantics-based method to cluster Chinese web search results. First, we employ the generalised suffix tree to extract the longest common substrings (LCSs) from search snippets. Second, we use the HowNet to calculate the similarities of the words derived from the LCSs, and extract the most representative features by constructing the vocabulary chain. Third, we construct a vector of text features and calculate snippets’ semantic similarities. Finally, we improve the Chameleon algorithm to cluster snippets. Extensive experimental results have shown that the proposed algorithm has outperformed over the suffix tree clustering method and other traditional clustering methods.