Search

Tag ranking has emerged as an important research topic recently due to its potential application on web image search. Existing tag relevance ranking approaches mainly rank the tags according to their relevance levels with respect to a given image. Nonetheless, such algorithms heavily rely on the large-scale image dataset and the proper similarity measurement to retrieve semantic relevant images with multi-labels. In contrast to the existing tag relevance ranking algorithms, in this paper, we propose a novel tag saliency ranking scheme, which aims to automatically rank the tags associated with a given image according to their saliency to the image content. To this end, this paper presents an integrated framework for tag saliency ranking, which combines both visual attention model and multi-instance learning to investigate the saliency ranking order information of tags with respect to the given image. Specifically, tags annotated on the image-level are propagated to the region-level via an efficient multi-instance learning algorithm firstly; then, visual attention model is employed to measure the importance of regions in the given image. Finally, tags are ranked according to the saliency values of the corresponding regions. Experiments conducted on the COREL and MSRC image datasets demonstrate the effectiveness and efficiency of the proposed framework.

Music Information Research (MIR) requires access to real musical content in order to test the efficiency and effectiveness of its methods as well as to compare developed methodologies on common data. Existing datasets do not address the research direction of musical track popularity that has recently received considerate attention. Moreover, sources of musical popularity do not provide easily manageable data and no standardised dataset exists for musical popularity research. To address these issues the Track Popularity Dataset (TPD) was created in a previous work. TPD provided (a) different sources of popularity definition ranging from 2004 to 2014, (b) mapping between different track/ author/ album identification spaces allowing use of different popularity sources, (c) information on the remaining, non popular, tracks of an album with a popular track, (d) contextual similarity between tracks and (e) ready for MIR use extracted features for both popular and non-popular audio tracks. This paper extends the TPD by (a) adding more readily computed features, (b) proposing feature & similarity definitions on popularity trends, (c) formulating common data mining scenarios on tracks’ popularity and (d) presenting respective promising results.

Recently, as location-based social networks (LBSNs) rapidly grow, general users utilize point-of-interest recommender systems to discover attractive locations. Most existing POI recommendation algorithms always employ the check-in data and rich contextual information (e.g., geographical information and users social network information) of users to learn their preference on POIs. Unfortunately, these studies generally suffer from two major limitations: (1) when modeling geographical influence, users personalized behavior differences are ignored; (2) when modeling the users social influence, the implicit social influence is seldom exploited. In this paper, we propose a novel POI recommendation approach called GeoEISo. GeoEISo achieves three key goals in this work. (1) We develop a kernel estimation method with a self-adaptive kernel bandwidth to model the geographical influence between POIs. (2) We use the Gaussian radial basis kernel function based support vector regression (SVR) model to predict explicit trust values between users, and then devise a novel trust-based recommendation model to simultaneously incorporate both the explicit and implicit social trust information into the process of POI recommendation. (3) We develop a unified geo-social framework which combines users preference on a POI with the geographical influence as well as social correlations. Experimental results on two real-world datasets collected from Foursquare show that GeoEISo provides significantly superior performances compared to other state-of-the-art POI recommendation models.

Search in social network is continuously being expanded to enhance user experience. Besides basic textual retrieval, users can also search based on features such as spatial proximity, temporal freshness and/or social closeness. To efficiently process each advanced query type, customized indexing mechanisms have been developed. However, such mechanisms only perform well for the query types that they were designed for; moreover, they are not readily adaptable to support other query types. In this paper, we propose an interval-at-a-time (IAAT) framework as a first attempt to provide a one-size-fits-all solution to social media retrieval with spatial, temporal and social constraints. In addition, the algorithm relies on inverted index only, which makes it compatible with conventional search engines. The inverted lists are sorted by document id and the insertion is very fast because only append operation is involved. Experiments conducted on two large-scale Twitter datasets show that though IAAT is a unified strategy, it performs better than most of the state-of-the-art customized solutions in a variety of query types.

Recommender system (RS) has become an active research area driven by the enormous industrial demands. Meanwhile, with the rapid development of microblogging system, various kinds of social data are available, which provide opportunities as well as challenges for traditional RSs. In this paper, we introduce the social recommendation (SR) problem utilizing microblogging data. We study this problem via multi-view user preference learning. Specifically, we first model user preference by learning a low-dimensional common representation of multi-view information including rating information, social relations, item side information, tagging information, and then recommend items based on the learnt user preference. We also develop an efficient alternating direction method of multipliers (ADMM) scheme to solve the proposed model. We empirically evaluate our approach using two real world datasets to demonstrate the significant improvement of our proposed approach against the state-of-the-art algorithms.

Search engine query recommendation based on mining query logs has been considered as an important and useful method of facilitating users to retrieve information. However, the log data evolves quickly. Existing query recommendation approaches have to rebuild the models when new log data arrive. In this paper, we extend the query ranking model (QRM) proposed in our previous work (Wang et al., 2015) [1] to an adaptive model in which new coming log data is incrementally added, so that the recommendation model is kept up-to-date. The experimental results have demonstrated that the proposed incremental query ranking model (IQRM) is able to recommend queries more efficiently than re-building QRM on evolving log data without losing accuracy.

WOS: 000382794500034, The basic idea behind privacy-preserving collaborative filtering schemes is to prevent data collectors from deriving the actual rating values and the rated items. Different data perturbation methods have been proposed to protect individual privacy. Due to different privacy concerns, users might disguise their data variably to meet their own privacy concerns. In addition to reconstructing the true rating values, data collectors might try to reconstruct the rated items. In this paper, our goal is to reconstruct the rated items with the help of auxiliary information when users mask their confidential data inconsistently in privacy-preserving prediction systems. We first need to estimate the number of the rated items. Then we have to predict the rated items. To do so, we first use existing methods to eliminate noise from the disguised data. We improve our predictions by utilizing the auxiliary information. Our real data-based empirical outcomes show that our proposed approaches are able to reconstruct the rated items with decent accuracy in spite of variable data masking, TUBITAK [113E262], This work is supported by the Grant 113E262 from TUBITAK.

Image tag-ranking, the task to sort tags based on their relevance to the related images, has become a hot topic in the field of multimedia. Most existing methods do not incorporate the tag-ranking order information into the models, which is actually very important to solve the issue of image tag-ranking. In this paper, by taking advantage of such important information, we propose a novel model which uses images with ranked tag lists as its supervision information. In the proposed method, each ranked tag list is decomposed into a number of image-tag pairs, all of which are pooled together for training a scoring function. With this pairwise supervision, the model is able to capture the intrinsic ranking structures. In addition, unsupervised data, namely images with unranked tag lists, is also integrated for digging the binary order: relevant or irrelevant. By leveraging both the pairwise supervision and unsupervised structural information, our model sufficiently exploits the tag relevance to images as well as the ranking structures of tag lists. Extensive experiments are conducted on both image tag-ranking and tag-based image search with three benchmark datasets: SUNAttribute, Labelme and MSRC, demonstrating the effectiveness of the proposed model.

Query recommendation technology is of great importance for search engines, because it can assist users to find the information they require. Many query recommendation algorithms have been proposed, but they all aim to recommend similar queries and cannot guarantee the usefulness of the recommended queries. In this paper, we argue that it is more important to recommend high-utility queries, i.e., queries that would induce users to search for more useful information. For this purpose, we propose a query-recommending model to rank candidate queries according to their utilities and to recommend those that are useful to users. The query-recommending model ranks a candidate query by assessing the joint probability that the query is selected by the user, that the obtained search results are subsequently clicked by the user, and that the clicked search results ultimately satisfy the user?s information need. Three utilities were defined to solve the model: query-level utility, representing the attractiveness of a query to the user; perceived utility, measuring the user?s probability of clicking on the search results; and posterior utility, measuring the useful information obtained by the user from the clicked search results. The methods that were used to compute these three utilities from the query log data are presented. The experimental results that were obtained by using real query log data demonstrated that the proposed query-recommending model outperformed six other baseline methods in generating more useful recommendations.

As video data generated by users boom continuously, making sense of large scale data archives is considered as a critical challenge for data management. Most existing learning techniques that extract signal-level contents from video data struggle to scale due to efficiency limits. With the development of pervasive positioning techniques, discovering hot topics from multimedia data by their geographical tags has become practical: videos taken by advanced cameras are associated with GPS locations, and geo-tagged videos from YouTube can be identified by their associated GPS locations on Google Maps. It enables us to know the cultures, scenes, and human behaviors from videos based on their spatio-temporal distributions. However, meaningful topic discovery requires an efficient clustering approach, through which coherent topics can be detected according to particular geographical regions without out-of-focus effects. To handle this problem, this paper presents a filter-refinement framework to discover hot topics corresponding to geographical dense regions, and then introduces two novel metrics to refine unbounded hot regions, together with a heuristic method for setting rational thresholds on these metrics. The results of extensive experiments prove that hot topics can be efficiently discovered by our framework, and more compact topics can be achieved after using the novel metrics.

Query difficulty estimation predicts the performance of the search result of the given query. It is a powerful tool for multimedia retrieval and receives increasing attention. It can guide the pseudo relevance feedback to rerank the image search results and re-write the query by suggesting "easy" alternatives to obtain better search results. Many techniques to estimate the query difficulty have been proposed in the textual information retrieval, but directly employing them for image search will result in poor performance. That is because image query is more complex with spatial or structural information, and the well-known semantic gap induces extra burdens for accurate estimations. In this paper, we propose a query difficulty estimation approach by analyzing the top ranked images obtained by ad hoc retrieval models. Specifically, we seamlessly integrate the language model based clarity score, the spatial consistency of local descriptors and the appearance consistency of global features. Experimental results demonstrate that the query difficulty estimated by the proposed algorithm correlates well with the actual retrieval performance. Two applications of query difficulty estimation, namely guided pseudo relevance feedback (GPRF) and selective query refinement (SQR), are also proposed from both system and user perspectives. Experimental results show that both strategies further boost the retrieval performance. © 2012 Elsevier B.V..

Search engines result pages (SERPs) for a specific query are constructed according to several mechanisms. One of them consists in ranking Web pages regarding their importance, regardless of their semantic. Indeed, relevance to a query is not enough to provide high quality results, and popularity is used to arbitrate between equally relevant Web pages. The most well-known algorithm that ranks Web pages according to their popularity is the PageRank. The term Webspam was coined to denotes Web pages created with the only purpose of fooling ranking algorithms such as the PageRank. Indeed, the goal of Webspam is to promote a target page by increasing its rank. It is an important issue for Web search engines to spot and discard Webspam to provide their users with a nonbiased list of results. Webspam techniques are evolving constantly to remain efficient but most of the time they still consist in creating a specific linking architecture around the target page to increase its rank. In this paper we propose to study the effects of node aggregation on the well-known ranking algorithm of Google (the PageRank) in the presence of Webspam. Our node aggregation methods have the purpose to construct clusters of nodes that are considered as a sole node in the PageRank computation. Since the Web graph is way to big to apply classic clustering techniques, we present four lightweight aggregation techniques suitable for its size. Experimental results on the WEBSPAM-UK2007 dataset show the interest of the approach, which is moreover confirmed by statistical evidence.

COSMOS-7 models semantic content in MPEG-7 such as objects and events and their spatio-temporality. When a user queries a COSMOS-7 model, the output is usually presented as a sequence of relevant, albeit unranked, video segments through which the user must sift. In this paper, we report how we use Self-Organising Neural Networks (SONNs) to cluster and rank the video segments through consideration of user preferences and knowledge gained from usage of the same content by similar users and of similar content by the same user.