A Method based on One-class SVM for News Recommendation

Browsing is one of the most popular ways to gather information in database with hypertext structure. In the browsing, a user continuously searches nodes which include useful information for her/him. Her/his interests, then, often change while the browsing. We call this type of browsing “context-sensitive browsing” in order to distinguish it from browsing with consistent interests. In this paper, we propose a method to filter the links in hypertext based on the user's browsing history. We assume that even when a user browses, following changeable interests without a clear task, the user's current interests are reflected in the content and order of nodes in the browsing history. The filtering method models user's current interests from the user's browsing history and puts the next choices in order of the nearness to the interests. We call the filtering method “context-sensitive filtering”. We have developed a browsing support system with this method for an encyclopedia in CD-ROM format. The results of an experimental evaluation, by real users, are also reported.

With the explosive growth of the number of news, how to help users efficiently and accurately find the news they are interested in from the mass of news has become more and more important. The traditional collaborative filtering news recommendation algorithm is based on user browsing records. However, the user's browsing history cannot directly indicate the user's preference for news. Meanwhile, the degree of correlation between user comments and news can intuitively reflect the degree of interest of users in news. Thus, a personalized news recommendation algorithm incorporating user comments is proposed. Firstly, the algorithm proposes a calculation method of user similarity, which integrates the influence of news popularity and the text characteristics of user comments. Secondly, we integrate the correlation between comments and news into the calculation of user's news preference, and measures the confidence of user's news browsing records according to the correlation level. Finally, we evaluate our approach with a large number of experiments. The experimental results show that the personalized news recommendation algorithm combined with user comments has significantly improved the recommendation effect compared with other algorithms.

Web browsers are implicitly trusted to handle a large amount of private user information. One such piece of private information, a user's browsing history, is maintained by browsers and is used to provide a popular usability feature: Web links are rendered using different styles depending on whether the URLs they point to have been visited or not. Unfortunately, this feature can be abused by malicious webpages in order to extract users’ browsing history. We present new browsing history sniffing attacks through two contention-based side channels which are new in this context: Last-level CPU cache contention, and GPU execution unit contention. The attacks are robust and can be executed successfully against the popular Chrome browser. Compared to prior work which uses the rendering performance as a side channel, our work achieves an attack rate increase of up to 30x. The new attacks are stealthier, because the side channels we use do not slow down the browser's rendering rate. In addition, we revisit the existing sniffing attacks based on the rendering performance side channel, and show how their attack rate can also be increased by a significant amount. Finally, we discuss the root cause of history sniffing attacks and point out solutions.

Preserving privacy of the user data is one of the important requirements in several practical applications. In recent times, the targeted advertisement is found a potential business model that the services provider can use it for revenue growth. As the advertising server (ad-server) tracks the user's activity through the user's browsing history or through the user's mobile application feature for future purposes, the privacy of user's data could fall in hand of many parties without the knowledge of the user. In this paper, a scheme for privacy preserving targeted online advertisement is presented. The proposed scheme uses homomorphic encryption aiming at extracting keywords from web pages in real-time when a user is browsing web pages. A user can avail the services privately from the ad-server by perform the computation on encrypted data stored in the server. The proposed scheme is compared with some related schemes and the experimental results show that the proposed scheme is efficient and practical

User data is the primary input of digital advertising, fueling the free Internet as we know it. As a result, web companies invest a lot in elaborate tracking mechanisms to acquire user data that can sell to data markets and advertisers. However, with same-origin policy, and cookies as a primary identification mechanism on the web, each tracker knows the same user with a different ID. To mitigate this, Cookie Synchronization (CSync) came to the rescue, facilitating an information sharing channel between third parties that may or not have direct access to the website the user visits. In the background, with CSync, they merge user data they own, but also reconstruct a user's browsing history, bypassing the same origin policy. In this paper, we perform a first to our knowledge in-depth study of CSync in the wild, using a year-long weblog from 850 real mobile users. Through our study, we aim to understand the characteristics of the CSync protocol and the impact it has on web users' privacy. For this, we design and implement CONRAD, a holistic mechanism to detect CSync events at real time, and the privacy loss on the user side, even when the synced IDs are obfuscated. Using CONRAD, we find that 97% of the regular web users are exposed to CSync: most of them within the first week of their browsing, and the median userID gets leaked, on average, to 3.5 different domains. Finally, we see that CSync increases the number of domains that track the user by a factor of 6.75.

Classifying web users in a personalised search setup is cumbersome due the very nature of dynamism in user browsing history. This fluctuating nature of user behaviour and user interest shall be well interpreted within a fuzzy setting. Prior to analysing user behaviour, nature of user interests has to be collected. This work proposes a fuzzy based user classification model to suit a personalised web search environment. The user browsing data is collected using an established customised browser designed to suit personalisation. The data are fuzzified and fuzzy rules are generated by applying decision trees. Using fuzzy rules, the search pages are labelled to aid grouping of user search interests. Evaluation of the proposed approach proves to be better when compared with Bayesian classifier.

User interest and behavior modeling is a critical step in online digital advertising. On the one hand, user interests directly impact their response and actions to the displayed advertisement (Ad). On the other hand, user interests can further help determine the probability of an Ad viewer becoming a buying customer. To date, existing methods for Ad click prediction, or click-through rate prediction, mainly consider representing users as a static feature set and train machine learning classifiers to predict clicks. Such approaches do not consider temporal variance and changes in user behaviors, and solely rely on given features for learning. In this paper, we propose two deep learning-based frameworks, {hbox {LSTM}}_{mathrm{cp}} and {hbox {LSTM}}_{mathrm{ip}}, for user click prediction and user interest modeling. Our goal is to accurately predict (1) the probability of a user clicking on an Ad and (2) the probability of a user clicking a specific type of Ad campaign. To achieve the goal, we collect page information displayed to the users as a temporal sequence and use long short-term memory (LSTM) network to learn features that represents user interests as latent features. Experiments and comparisons on real-world data show that, compared to existing static set-based approaches, considering sequences and temporal variance of user requests results in improvements in user Ad response prediction and campaign specific user Ad click prediction.

A huge number of display advertising campaigns are launched every day by advertisers in order to promote their products or services. The main objective of each advertiser is to display ads to specific groups of users, i.e. users who meet specific criteria or their interests are related to the promoted products or services. Audience expansion, also known as audience look-alike targeting, is one of the major display advertising techniques that helps advertisers to discover audiences with similar attributes to a target audience who is interested in advertisers' products or services. In this paper, we present different audience expansion schemes able to identify users with similar browsing interests to those of the seed users provided by the advertiser. The proposed audience expansion schemes are based on different unsupervised representation models that are able to capture the interests of the users according to their browsing history. We have conducted an extensive empirical study on a real data collected from an advertising platform to analyse the effectiveness of the proposed schemes to expand the audiences of five different advertisers.

As the rapid growth of digital scientific and technical literatures, the scientific researchers need personalized retrieval to satisfy their research interests urgently. Unlike previous methods that just use simple keyword matching, we strengthen the semantic information of scientific literature by merging metadata such as title, keywords, abstract and citation in this dissertation. Then we use vector space model with tf-idf value of each term to model each scientific literature. We structure user's interest model by subject term vector with different weight. Different weight means different concern degree to each subject term by user. To emphasize the quality of recommended literature, we not only calculate the similarity between user's research interest and literatures, but also consider the total times cited of recommended literatures. As experimental dataset, we collect literatures from Web of Science under the topic of “pressure sensor”. The experimental results show that this recommendation method could well indentify user's research interest. So, it can improve efficiency of user literature retrieval and increase the accuracy of the literature recommendation.

We propose a method that helps users to understand their own interests by extracting terms from selected link texts and generating a new browsing history, and arranging those terms and Web-page icons on to the user's interest space in chronological order. We have implemented a prototype system based on this method. The system's performance was evaluated in two experiments, which revealed that (1) Japanese interest terms can be extracted from the user's selected link texts, (2) the interest-space browser displays the user's interest space on browsing the Web, and (3) the users can use this system to better understand their own interests, recall their past, and reorganize previously browsed Web pages.

Recently, exploiting a knowledge graph (KG) to enrich the semantic representation of a news article have been proven to be effective for news recommendation. These solutions focus on the representation learning for news articles with additional information in the knowledge graph, where the user representations are mainly derived based on these news representations later. However, different users would hold different interests on the same news article. In other words, directly identifying the entities relevant to the user's interest and deriving the resultant user representation could enable a better news recommendation and explanation.

Search engines usually return relevant sorted results based on the keywords. Because of the lack of considering the user's current search interest and intention, this kind of strategy may not meet users' personalized search requirements. In order to retrieve results associated with the user's current search interest, existing researches integrate the user's interest information into the search process and ranking, so as to make the search results fit to the user's current search target more better. Different from the unified model of interest which most current related researches used, we propose a model of combing user's interests for the purpose of ranking in the light of different query types. This paper first classified queries into several types based on the current keywords and the sources of user's interests, such as short-term interest in user clicks, and long-term interest in user's search history, search logs of the search system etc. Then based on the sources of interests, we established a model of combination of interests which based on the artificial neural network and was trained by related approaches and data sets. Depending on different query types, we combined these sources of user's interests that the most best reflecting the user's current search intention to enhance the user's search experience. Comparing with the existing model of single interest, the results of the experiments show that our model can obtain better ranking on the results.

Online news are being spread through social media by news agencies to encourage people to read news from their site. After users log in to their site, users will continue to read the news if it is relevant to their personalized news recommendation. However, nowadays, a personalized recommendation could be provided to users if the site has records of users' browsing history, and users must have logged in to the news site. This could be problematic if the readers are anonymous users, and the site does not have enough browsing history records of these users. It will be difficult for news agencies to increase the number of daily readers on their site, especially if they have to compete with other agencies for providing a good personalized recommendation. Therefore, a personalized recommendation system is necessary for anonymous users who only visit the news site occasionally so that the website can recommend the news they prefer to read by implementing CSR Sparse Matrix Vector Multiplication and proximity processing as sparse matrix applied method. The Bayesian framework for user interest will also be implemented with the modification in its term. This research proved that personalized news recommendations could still be provided to readers with limited knowledge of the users' searching history.

A major shortcoming of content-based approaches exists in the representation of the user model. Content-based approaches often employ term vectors to represent each user's interest. In doing so, they ignore the semantic relations between terms of the vector space model in which indexed terms are not orthogonal and often have semantic relatedness between one another. In this paper, we improve the representation of a user model during building user model in content-based approaches by performing these steps. First is the domain concept filtering in which concepts and items of interests are compared to the domain ontology to check the relevant items to our domain using ontology based semantic similarity. Second, is incorporating semantic content into the term vectors. We use word definitions and relations provided by WordNet to perform word sense disambiguation and employ domain-specific concepts as category labels for the semantically enhanced user models. The implicit information pertaining to the user behavior was extracted from click stream data or web usage sessions captured within the web server logs. Also, our proposed approach aims to update user model, we should analysis user's history query keywords. For a certain keyword, we extract the words which have the semantic relationships with the keyword and add them into the user interest model as nodes according to semantic relationships in the WordNet. Keywords-User model; Domain ontology; Semantic Similarity; Wordnet.

How do users interact with algorithm recommender systems? The interaction of users, algorithms, and performance