Socially-Aware Recommendations In Smart Conferences

The current global proliferation of events such as academic conferences, symposia and workshops has introduced high enormities of academic resources. Academic conferences are very important, due to the fact that they foster potential collaborations, enhance social awareness and improve familiarization and interaction among participants. During conference events, participants encounter different academic resources. These academic resources usually include different participants, scholarly/research papers and presentation sessions. Due to the high patronage of current academic conferences, the number of these resources has become very vast. This has introduced the problem of information overload regarding academic resources, making it difficult for participants to easily acquire the right resources they need.Consequently, the reduction of information overload at academic conferences is an important research direction which requires immediate and urgent attention. It has therefore become very necessary to address such problems so that conference participants can acquire the right, relevant and necessary information they need for effective social and personal learning. Recommender systems, which have become a ubiquitous and important feature in our society, can be used to address information overload problems at academic conferences. In comparison to traditional academic conferences, smart conferences enable participants to interact with each other and acquire social recommendations through mobile devices (smartphones).In addition to utilizing user interests and items, we innovatively integrate and apply social properties (social ties and degree centrality) of users (participants) in smart conferences, and address three main problems:(a) How can we socially recommend scholarly/research papers to participants to reduce the difficulties they face in acquiring scholarly/research papers of interest?(b) How can we socially recommend presentation session venues and environments to participants so that they can attend the right presentation session(s) in accordance to common research interests with a presenter?(c) How can we socially recommend unknown participants to each other based on their research interests to enhance research collaboration and interaction?To solve the first problem of this dissertation, we initially analyze the social tie property of users and design our first novel recommender algorithm called Socially-Aware Recommendation of Scholarly Papers (SARSP). SARSP computes the social ties among conference participants using their physical contact durations and contact frequencies. The innovative computation of social ties in SARSP signifies both weak and strong ties among conference participants and this enables active participants to recommend their scholarly/research papers to other participants with strong ties in accordance to a determined social tie threshold. SARSP further computes the folksonomies among conference participants and utilizes a normalization procedure to generate group profiles of the conference participants using the computed folksonomies. The novelty involved with folksonomies and group profile generation enables active participants to recommend their scholarly/research papers to other participants in a group (more participants at a time) rather than individuals based on strong research interests in terms of scholarly/research papers. Through a series of benchmarking experiments involving a relevant dataset, we evaluate the recommendation outputs of SARSP, namely:social tie recommendation and social group recommendation using recall, precision and f-measure evaluation metrics. Our experimental results innovatively show that SARSP outperforms other compared state-of-the-art methods in terms of the above metrics. Furthermore, due to the utilization of innovative social information through social properties, SARSP favorably reduces cold start and data sparsity challenges.To solve the second problem of this dissertation, we initially analyze the social tie and degree centrality properties of users in relation to communities and context. Our second novel recommender algorithm called Socially-Aware Recommendation of Venues and Environments (SARVE), computes the social ties among participants and presenters using their physical contact durations and contact frequencies. SARVE also computes the degree centrality of presenters to determine their popularity. The innovative computations of social ties and degree centrality in our recommender algorithm indicates both weak and strong ties among participants and presenters. These computations enable the recommendation of presentations sessions to participants based on strong ties with presenter(s) in accordance to a determined social tie threshold and the popularity level of presenters. SARVE further computes the correlation among participants and presenters in terms of research interests and uses a suitable model to match contextual relationships and generate effective recommendations. Through a series of benchmarking experiments involving a relevant dataset, we evaluate the recommendation outputs of SARVE, namely:social relations recommendation and social context recommendation using precision, recall and f-measure evaluation metrics. Our experimental results innovatively show that, in comparison to other compared state-of-the-art methods, SARVE performs better in terms of the utilized metrics. Additionally, due to the utilization of innovative social information through social properties and context, SARVE satisfactorily alleviates cold start and data sparsity challenges.To solve the third problem of this dissertation, we initially analyze the estimated social ties and personality traits of users. Our third novel recommender algorithm called Socially and Personality Aware Recommendation of Participants (SPARP), computes the estimated (accurate) social ties among participants using their physical contact durations and contact frequencies. In order to improve academic/research collaboration, SPARP innovatively recommends strangers/unknown (weak tied) participants to each other based on a weighted hybrid combination of strong personality traits and estimated social ties. Through a series of benchmarking experiments, we evaluate the recommendation output of SPARP, namely: merging similarity coefficients with different weight parameters, using accuracy, Mean Absolute Error (MAE) and Normalized Mean Absolute Error (NMAE) evaluation metrics. In comparison to other compared state-of-the-art methods, our experimental results innovatively reveal that SPARP performs better in terms of the employed metrics. Moreover, due to the utilization of an innovative weighted hybrid procedure that combines social information and personality, SPARP positively avoids cold start and data sparsity challenges.In order to corroborate recommendation quality and accuracy through the encouraging and innovative experimental results we achieved, our social recommendation methods can be implemented and used in current smart conference environments, where each participant is equipped with a smartphone that has standard specifications.