Yuanxi Li

Intelligent Social Media Indexing and Sharing Using an Adaptive Indexing Search Engine

Effective sharing of diverse social media is often inhibited by limitations in their search and discovery mechanisms, which are particularly restrictive for media that do not lend themselves to automatic processing or indexing. Here, we present the structure and mechanism of an adaptive search engine which is designed to overcome such limitations. The basic framework of the adaptive search engine is to capture human judgment in the course of normal usage from user queries in order to develop semantic indexes which link search terms to media objects semantics. This approach is particularly effective for the retrieval of multimedia objects, such as images, sounds, and videos, where a direct analysis of the object features does not allow them to be linked to search terms, for example, nontextual/icon-based search, deep semantic search, or when search terms are unknown at the time the media repository is built. An adaptive search architecture is presented to enable the index to evolve with respect to user feedback, while a randomized query-processing technique guarantees avoiding local minima and allows the meaningful indexing of new media objects and new terms. The present adaptive search engine allows for the efficient community creation and updating of social media indexes, which is able to instill and propagate deep knowledge into social media concerning the advanced search and usage of media resources. Experiments with various relevance distribution settings have shown efficient convergence of such indexes, which enable intelligent search and sharing of social media resources that are otherwise hard to discover.

Collective evolutionary concept distance based query expansion for effective web document retrieval

In this work several semantic approaches to concept-based query expansion and re-ranking schemes are studied and compared with different ontology-based expansion methods in web document search and retrieval. In particular, we focus on concept-based query expansion schemes where, in order to effectively increase the precision of web document retrieval and to decrease the users’ browsing time, the main goal is to quickly provide users with the most suitable query expansion. Two key tasks for query expansion in web document retrieval are to find the expansion candidates, as the closest concepts in web document domain, and to rank the expanded queries properly. The approach we propose aims at improving the expansion phase for better web document retrieval and precision. The basic idea is to measure the distance between candidate concepts using the PMING distance, a collaborative semantic proximity measure, i.e. a measure which can be computed using statistical results from a web search engine. Experiments show that the proposed technique can provide users with more satisfying expansion results and improve the quality of web document retrieval.

Set Similarity Measures for Images Based on Collective Knowledge

This work introduces a new class of group similarity where different measures are parameterized with respect to a basic similarity defined on the elements of the sets. Group similarity measures are of great interest for many application domains, since they can be used to evaluate similarity of objects in term of the similarity of the associated sets, for example in multimedia collaborative repositories where images, videos and other multimedia are annotated with meaningful tags whose semantics reflects the collective knowledge of a community of users. The group similarity classes are formally defined and their properties are described and discussed. Experimental results, obtained in the domain of images semantic similarity by using search engine based tag similarity, show the adequacy of the proposed approach in order to reflect the collective notion of semantic similarity.

Web-based similarity for emotion recognition in web objects

In this project we propose a new approach for emotion recognition using web-based similarity (e.g. confidence, PMI and PMING). We aim to extract basic emotions from short sentences with emotional content (e.g. news titles, tweets, captions), performing a web-based quantitative evaluation of semantic proximity between each word of the analyzed sentence and each emotion of a psychological model (e.g. Plutchik, Ekman, Lovheim). The phases of the extraction include: text preprocessing (tokenization, stop words, filtering), search engine automated query, HTML parsing of results (i.e. scraping), estimation of semantic proximity, ranking of emotions according to proximity measures. The main idea is that, since it is possible to generalize semantic similarity under the assumption that similar concepts co-occur in documents indexed in search engines, therefore also emotions can be generalized in the same way, through tags or terms that express them in a particular language, ranking emotions. Training results are compared to human evaluation, then additional comparative tests on results are performed, both for the global ranking correlation (e.g. Kendall, Spearman, Pearson) both for the evaluation of the emotion linked to each single word. Different from sentiment analysis, our approach works at a deeper level of abstraction, aiming to recognize specific emotions and not only the positive/negative sentiment, in order to predict emotions as semantic data.

Leveraging Zero Tail in Neighbourhood for Link Prediction

For link prediction, Common Neighbours (CN) ranking measures allow to discover quality links between nodes in a social network, assessing the likelihood of a new link based on the neighbours frontier of the already existing nodes. A zero rank value is often given to a large number of pairs of nodes, which have no common neighbours, that instead can be potentially good candidates for a quality assessment. With the aim of improving the quality of the ranking for link prediction, in this work we propose a general technique to evaluate the likelihood of a linkage, iteratively applying a given ranking measure to the Quasi-Common Neighbours (QCN) of the node pair, i.e. iteratively considering paths between nodes, which include more than one traversing step. Experiments held on a number of datasets already accepted in literature show that QCNAA, our QCN measure derived from the well know Adamic-Adar (AA), effectively improves the quality of link prediction methods, keeping the prediction capability of the original AA measure. This approach, being general and usable with any CN measure, has many different applications, e.g. trust management, terrorism prevention, disambiguation in co-authorship networks.

Context-based image semantic similarity

In this work we propose Context-based Image Similarity, a scheme for discovering and evaluating image similarity in terms of the associated groups of concepts. Several semantic proximity/similarity among image concepts and different concept ontology - WordNet Distance, Wikipedia Distance, Flickr Distance, Confidence, Normalized Google Distance (NGD), Pointwise Mutual Information (PMI) and PMING, have been considered as elementary metrics for the context. Comparing to Content Based Image Retrieval (CBIR), which measures the image content similarities by low level features, the proposed Context-based Image Similarity outperformed CBIR in measuring the deep concept similarity and relationship of images. Experimental results, obtained in the domain of images semantic similarity using search engine based tag similarity, show the adequacy of the proposed approach in order to reflect the collective notion of semantic similarity.

Clustering Facebook for Biased Context Extraction

Facebook comments and shared posts often convey human biases, which play a pivotal role in information spreading and content consumption, where short information can be quickly consumed, and later ruminated. Such bias is nevertheless at the basis of human-generated content, and being able to extract contexts that does not amplify but represent such a bias can be relevant to data mining and artificial intelligence, because it is what shapes the opinion of users through social media. Starting from the observation that a separation in topic clusters, i.e. sub-contexts, spontaneously occur if evaluated by human common sense, especially in particular domains e.g. politics, technology, this work introduces a process for automated context extraction by means of a class of path-based semantic similarity measures which, using third party knowledge e.g. WordNet, Wikipedia, can create a bag of words relating to relevant concepts present in Facebook comments to topic-related posts, thus reflecting the collective knowledge of a community of users. It is thus easy to create human-readable views e.g. word clouds, or structured information to be readable by machines for further learning or content explanation, e.g. augmenting information with time stamps of posts and comments. Experimental evidence, obtained by the domain of information security and technology over a sample of 9M3k page users, where previous comments serve as a use case for forthcoming users, shows that a simple clustering on frequency-based bag of words can identify the main context words contained in Facebook comments identifiable by human common sense. Group similarity measures are also of great interest for many application domains, since they can be used to evaluate similarity of objects in term of the similarity of the associated sets, can then be calculated on the extracted context words to reflect the collective notion of semantic similarity, providing additional insights on which to reason, e.g. in terms of cognitive factors and behavioral patterns.

A path-based model for emotion abstraction on facebook using sentiment analysis and taxonomy knowledge

Each term in a short text can potentially convey emotional meaning. Facebook comments and shared posts often convey human biases, which play a pivotal role in information spreading and content consumption. Such bias is at the basis of human-generated content, and capable of conveying contexts which shape the opinion of users through the social media flow of information. Starting from the observation that a separation in topic clusters, i.e. sub-contexts, spontaneously occur if evaluated by human common sense, this work introduces a process for automated extraction of sub-context in Facebook. Basing on emotional abstraction and valence, the automated extraction is exploited through a class of path-based semantic similarity measures and sentiment analysis. Experimental results are obtained using validated clustering techniques on such features, on the domain of information security, over a sample of over 9 million page users. An additional expert evaluation of clusters in tag clouds confirms that the proposed automated algorithm for emotional abstraction clusters Facebook comments compatibly with human common sense. The baseline methods rely on the robust notion of collective concept similarity.

Clustering Students Interactions in eLearning Systems for Group Elicitation

In this work we introduce a novel Learning Analytics approach to identify students’ communities. The introduction of Learning Management Systems in higher education requires the educators to plan their Learning Design (LD) process with the online scenario in mind. We examined the blended learning environment where this process takes place in the Virtual Learning Environment. This allows the educators to track most of the students’ individual activities, but the communications may be excluded from tracking since the students can use side communications channels, such as face-to-face communication, instant messaging and social network platforms. Our approach, using the student-system interactions histories, helps to discover hidden relationships among the students. The elicited information about students’ groupings and social interactions’ evolution over time can be used by educators to adapt and improve their LD process, to find associations between students’ social interactions and their academic performance, as well as to promote team-based learning.

Comparison of Different Ontology-Based Query Expansion Algorithms for Effective Image Retrieval

We study several semantic concept-based query expansion and re-ranking scheme and compare different ontology-based expansion methods in image search and retrieval. In particular, we exploit the two concept similarities of different concept expansion ontology-WordNet Similarity, Wikipedia Similarity. Furthermore, we compare the keywords semantic distance with the precision of image search results with query expansion according to different concept expansion algorithms. We also compare the image retrieval precision of searching with the expanded query and original plain query. Preliminary experiments have been able to demonstrate that the two proposed retrieval mechanism has the potential to outperform unaided approaches.