Using search engines to retrieve information has become an important part of people's daily lives. For most search engines, click information is an important factor in document ranking. As a result, some websites cheat to obtain a higher rank by fraudulently increasing clicks to their pages, which is referred to as "Click Spam". Based on an analysis of the features of fraudulent clicks, a novel automatic click spam detection approach is proposed in this paper, which consists of 1. modeling user sessions with a triple sequence, which, to the best of our knowledge, takes into account not only the user action but also the action objective and the time interval between actions for the first time; 2. using the user-session bipartite graph propagation algorithm to take advantage of cheating users to find more cheating sessions; and 3. using the pattern-session bipartite graph propagation algorithm to obtain cheating session patterns to achieve higher precision and recall of click spam detection. Experimental results based on a Chinese commercial search engine using real-world log data containing approximately 80 million user clicks per day show that 2.6% of all clicks were detected as spam with a precision of up to 97%.

Combating Web spam is one of the greatest challenges for Web search engines. State-of-the-art anti-spam techniques focus mainly on detecting varieties of spam strategies, such as content spamming and link-based spamming. Although these anti-spam approaches have had much success, they encounter problems when fighting against a continuous barrage of new types of spamming techniques. We attempt to solve the problem from a new perspective, by noticing that queries that are more likely to lead to spam pages/sites have the following characteristics: 1) they are popular or reflect heavy demands for search engine users and 2) there are usually few key resources or authoritative results for them. From these observations, we propose a novel method that is based on click-through data analysis by propagating the spamicity score iteratively between queries and URLs from a few seed pages/sites. Once we obtain the seed pages/sites, we use the link structure of the click-through bipartite graph to discover other pages/sites that are likely to be spam. Experiments show that our algorithm is both efficient and effective in detecting Web spam. Moreover, combining our method with some popular anti-spam techniques such as TrustRank achieves improvement compared with each technique taken individually.

Click-through behaviors are treated as invaluable sources of user feedback and they have been leveraged in several commercial search engines in recent years. However, estimating unbiased relevance is always a challenging task because of position bias. To solve this problem, many researchers have proposed a variety of assumptions to model click-through behaviors. Most of these models share a common examination hypothesis, which is that users examine search results from the top to the bottom. Nevertheless, this model cannot draw a complete picture of information-seeking behaviors. Many eye-tracking studies find that user interactions are not sequential but contain revisiting patterns. If a user clicks on a higher ranked document after having clicked on a lower-ranked one, we call this scenario a revisiting pattern, and we believe that the revisiting patterns are important signals regarding a user's click preferences. This paper incorporates revisiting behaviors into click models and introduces a novel click model named Temporal Hidden Click Model (THCM). This model dynamically models users' click behaviors with a temporal order. In our experiment, we collect over 115 million query sessions from a widely-used commercial search engine and then conduct a comparative analysis between our model and several state-of-the-art click models. The experimental results show that the THCM model achieves a significant improvement in the Normalized Discounted Cumulative Gain (NDCG), the click perplexity and click distributions metrics.

Combating Web spam has become one of the top challenges for Web search engines. State-of-the-art spam-detection techniques are usually designed for specific, known types of Web spam and are incapable of dealing with newly appearing spam types efficiently. With user-behavior analyses from Web access logs, a spam page-detection algorithm is proposed based on a learning scheme. The main contributions are the following. (1) User-visiting patterns of spam pages are studied, and a number of user-behavior features are proposed for separating Web spam pages from ordinary pages. (2) A novel spam-detection framework is proposed that can detect various kinds of Web spam, including newly appearing ones, with the help of the user-behavior analysis. Experiments on large-scale practical Web access log data show the effectiveness of the proposed features and the detection framework.

User behavior analysis has played an important role in Web information retrieval. Rare queries, whose frequencies are rather low, are usually ignored in existing studies due to the data sparseness. Little has been known about the mass of rare queries on either the information need or the user behavior. In this paper, we make an empirical study of users' behavior on rare queries using a large scale search log. Features concerning query, resource and post-query actions are analyzed, based on which we propose a practical categorization framework and obtain an overview of rare query composition. Further, we study the characteristics of several most commonly occurring types of rare queries, and suggest improving the search performance of them separately. This work gives more insights into understanding the long tail of queries and will be helpful for Web search in terms of rare queries.

Nowadays, user behavior analysis and collaborative filtering have drawn a large body of research in the machine learning community. The goal is either to enhance the user experience or discover useful information hidden in the data. In this article, we conduct extensive experiments on a Chinese input method data set, which keeps the word lists that users have used. Then, from the collaborative perspective, we aim to solve two tasks in natural language processing, that is, related word retrieval and new word detection. Motivated by the observation that two words are usually highly related to each other if they co-occur frequently in users’ records, we propose a novel semantic relatedness measure between words that takes both user behaviors and collaborative filtering into consideration. We utilize this measure to perform related word retrieval and new word detection tasks. Experimental results on both tasks indicate the applicability and effectiveness of our method.

Under different language contexts, people choose different terms or phrases to express their feelings and opinions. When a user is writing a paper or chatting with a friend, he/she applies a specific language model corresponding to the underlying goal. This paper presents a log-based study of analyzing the language models with specific goals. We exhibit the statistical information of terms and software programs, propose some methods to estimate the divergence of language models with specific user goals and measure the discrimination of these models. Experimental results show that the language models with different user goals have large divergence and different discrimination. These study conclusions can be applied to understand user needs and improve Human-Computer Interaction (HCI).

Ranking is an essential part of information retrieval(IR) tasks such as Web search. Nowadays there are hundreds of features for ranking. So learning to rank(LTR), an interdisciplinary field of IR and machine learning(ML), has attracted increasing attention. Those features used in the IR are not always independent from each other, hence the feature selection, an important issue in ML, should be paid attention to for LTR. However, the state-of-the-art LTR approaches merely analyze the connection among the features from the aspects of feature selection. In this paper, we propose a hierarchical feature selection strategy containing 2 phases for ranking and learn ranking functions. The experimental results show that ranking functions based on the selected feature subset significantly outperform the ones based on all features.

We consider the problem of detecting epidemic tendency by mining search logs. We propose an algorithm based on click-through information to select epidemic related queries/terms. We adopt linear regression to model epidemic occurrences and frequencies of epidemic related terms (ERTs) in search logs. The results show our algorithm is effective in finding ERTs which obtain a high correlation value with epidemic occurrences. We also find the proposed method performs better when combining different ERTs than using single ERT.

In this paper, we study on the reliability of search engine users using click-through data. We proposed a graph-based approach to evaluate user reliability according to how users click on search result lists. We tried to incorporate this measure of reliability into relevance feedback for improving ranking performances. Experimental results indicate that the proposed approach is both effective and applicable.

Mining feedback information from user click-through data is an important issue for modern Web retrieval systems in terms of architecture analysis, performance evaluation and algorithm optimization. For commercial search engines, user click-through data contains useful information as well as large amount of inevitable noises. This paper proposes an approach to recognize reliable and meaningful user clicks (referred to as Relevant Clicks, RCs) in click-through data. By modeling user click-through behavior on search result lists, we propose several features to separate RCs from click noises. A learning algorithm is presented to estimate the quality of user clicks. Experimental results on large scale dataset show that: 1) our model effectively identifies RCs in noisy click-through data; 2) Different from previous click-through analysis efforts, our approach works well for both hot queries and long-tail queries.

How many pages on the Web will be accessed by Web users? This is an interesting question for both Web scientists and industry engineers. To answer this question, User Access Web (UA Web) is described and studied in this paper. With analysis on large scale Web users’ access logs, a sampling procedure is proposed to reduce the bias, and the near-uniform random pages are sampled from the UA Web applying search engine interface and Monte Carlo methods. Experimental results on about 675 million user log entries reveal some properties of the UA Web and the indices of four search engines, e.g. power law distribution, average length of pages, index size of search engines, properties of static and dynamic pages, etc.

News topics, which are constructed from news stories using the techniques of Topic Detection and Tracking (TDT), bring convenience to users who intend to see what is going on through the Internet. However, it is almost impossible to view all the generated topics, because of the large amount. So it will be helpful if all topics are ranked and the top ones, which are both timely and important, can be viewed with high priority. Generally, topic ranking is determined by two primary factors. One is how frequently and recently a topic is reported by the media; the other is how much attention users pay to it. Both media focus and user attention varies as time goes on, so the effect of time on topic ranking has already been included. However, inconsistency exists between both factors. In this paper, an automatic online news topic ranking algorithm is proposed based on inconsistency analysis between media focus and user attention. News stories are organized into topics, which are ranked in terms of both media focus and user attention. Experiments performed on practical Web datasets show that the topic ranking result reflects the influence of time, the media and users. The main contributions of this paper are as follows. First, we present the quantitative measure of the inconsistency between media focus and user attention, which provides a basis for topic ranking and an experimental evidence to show that there is a gap between what the media provide and what users view. Second, to the best of our knowledge, it is the first attempt to synthesize the two factors into one algorithm for automatic online topic ranking.

News Topics are related to a set of keywords or keyphrases. Topic keyphrases briefly describe the key content of topics and help users decide whether to do further reading about them. Moreover, keyphrases of a news topic can be considered as a cluster of related terms, which provides term relationship information that can be integrated into information retrieval models. In this paper, an automatic online news topic keyphrase extraction system is proposed. News stories are organized into topics. Keyword candidates are firstly extracted from single news stories and filtered with topic information. Then a phrase identification process combines keywords into phrases using position information. Finally, the phrases are ranked and top ones are selected as topic keyphrases. Experiments performed on practical Web datasets show that the proposed system works effectively, with a performance of precision=70.61% and recall=67.94%.

Combating Web spam has become one of the top challenges for Web search engines. State-of-the-art spam detection techniques are usually designed for specific known types of Web spam and are incapable and inefficient for recently-appeared spam. With user behavior analyses into Web access logs, we propose a spam page detection algorithm based on Bayes learning. Preliminary experiments on Web access data collected by a commercial Web site (containing over 2.74 billion user clicks in 2 months) show the effectiveness of the proposed detection framework and algorithm.

In many cases, rather than a keyword search, people intend to see what is going on through the Internet. Then the integrated comprehensive information on news topics is necessary, which we called news issues, including the background, history, current progress, different opinions and discussions, etc. Traditionally, news issues are manually generated by website editors. It is quite a time-consuming hard work, and hence real-time update is difficult to perform. In this paper, a three-step automatic online algorithm for news issue construction is proposed. The first step is a topic detection process, in which newly appearing stories are clustered into new topic candidates. The second step is a topic tracking process, where those candidates are compared with previous topics, either merged into old ones or generating a new one. In the final step, news issues are constructed by the combination of related topics and updated by the insertion of new topics. An automatic online news issue construction process under practical Web circumstances is simulated to perform news issue construction experiments. F-measure of the best results is either above (topic detection) or close to (topic detection and tracking) 90%. Four news issue construction results are successfully generated in different time granularities: one meets the needs like "what's new", and the other three will answer questions like "what's hot" or "what's going on". Through the proposed algorithm, news issues can be effectively and automatically constructed with real-time update, and lots of human efforts will be released from tedious manual work.

Performance evaluation is an important issue in Web search engine researches. Traditional evaluation methods rely on much human efforts and are therefore quite time-consuming. With click-through data analysis, we proposed an automatic search engine performance evaluation method. This method generates navigational type query topics and answers automatically based on search users. querying and clicking behavior. Experimental results based on a commercial Chinese search engine's user logs show that the automatically method gets a similar evaluation result with traditional assessor-based ones.

Abstract Understanding what kinds of Web pages are the most useful for Web search engine users is a critical task in Web information retrieval (IR). Most previous works used hyperlink analysis algorithms to solve this problem. However, little research has been focused on query‐independent Web data cleansing for Web IR. In this paper, we first provide analysis of the differences between retrieval target pages and ordinary ones based on more than 30 million Web pages obtained from both the Text Retrieval Conference (TREC) and a widely used Chinese search engine, SOGOU (www.sogou.com). We further propose a learning‐based data cleansing algorithm for reducing Web pages that are unlikely to be useful for user requests. We found that there exists a large proportion of low‐quality Web pages in both the English and the Chinese Web page corpus, and retrieval target pages can be identified using query‐independent features and cleansing algorithms. The experimental results showed that our algorithm is effective in reducing a large portion of Web pages with a small loss in retrieval target pages. It makes it possible for Web IR tools to meet a large fraction of users' needs with only a small part of pages on the Web. These results may help Web search engines make better use of their limited storage and computation resources to improve search performance.