Social Business Intelligence (SBI) enables companies to capture strategic information from public social networks. Contrary to traditional Business Intelligence (BI), SBI has to face the high dynamicity of both the social network’s contents and the company’s analytical requests, as well as the enormous amount of noisy data. Effective exploitation of these continuous sources of data requires efficient processing of the streamed data to be semantically shaped into insightful facts. In this paper, we propose a multidimensional formalism to represent and evaluate social indicators directly from fact streams derived in turn from social network data. This formalism relies on two main aspects: the semantic representation of facts via Linked Open Data and the support of OLAP-like multidimensional analysis models. Contrary to traditional BI formalisms, we start the process by modeling the required social indicators according to the strategic goals of the company. From these specifications, all the required fact streams are modeled and deployed to trace the indicators. The main advantages of this approach are the easy definition of on-demand social indicators, and the treatment of changing dimensions and metrics through streamed facts. We demonstrate its usefulness by introducing a real scenario user case in the automotive sector.

Background: Recent work in social network analysis has shown the usefulness of analysing and predicting outcomes from user-generated data in the context of Public Health Surveillance (PHS). Most of the proposals have focused on dealing with static datasets gathered from social networks, which are processed and mined off-line. However, little work has been done on providing a general framework to analyse the highly dynamic data of social networks from a multidimensional perspective. In this paper, we claim that such a framework is crucial for including social data in PHS systems. Methods: We propose a dynamic multidimensional approach to deal with social data streams. In this approach, dynamic dimensions are continuously updated by applying unsupervised text mining methods. More specifically, we analyse the semantics and temporal patterns in posts for identifying relevant events, topics and users. We also define quality metrics to detect relevant user profiles. In this way, the incoming data can be further filtered to cope with the goals of PHS systems. Results: We have evaluated our approach over a long-term stream of Twitter. We show how the proposed quality metrics allow us to filter out the users that are out-of-domain as well as those with low quality in their messages. We also explain how specific user profiles can be identified through their descriptions. Finally, we illustrate how the proposed multidimensional model can be used to identify main events and topics, as well as to analyse their audience and impact. Conclusions: The results show that the proposed dynamic multidimensional model is able to identify relevant events and topics and analyse them from different perspectives, which is especially useful for PHS systems.

Social media platforms have become a new source of useful information for companies. Ensuring the business value of social media first requires an analysis of the quality of the relevant data and then the development of practical business intelligence solutions. This paper aims at building high-quality datasets for social business intelligence (SoBI). The proposed method offers an integrated and dynamic approach to identify the relevant quality metrics for each analysis domain. This method employs a novel multidimensional data model for the construction of cubes with impact measures for various quality metrics. In this model, quality metrics and indicators are organized in two main axes. The first one concerns the kind of facts to be extracted, namely: posts, users, and topics. The second axis refers to the quality perspectives to be assessed, namely: credibility, reputation, usefulness, and completeness. Additionally, quality cubes include a user-role dimension so that quality metrics can be evaluated in terms of the user business roles. To demonstrate the usefulness of this approach, the authors have applied their method to two separate domains: automotive business and natural disasters management. Results show that the trade-off between quantity and quality for social media data is focused on a small percentage of relevant users. Thus, data filtering can be easily performed by simply ranking the posts according to the quality metrics identified with the proposed method. As far as the authors know, this is the first approach that integrates both the extraction of analytical facts and the assessment of social media data quality in the same framework.

This paper presents a novel author profiling method specially aimed at classifying social network users into the multidimensional perspectives for social business intelligence (SBI) applications. In this scenario, being the user profiles defined on demand for each particular SBI application, we cannot assume the existence of labelled datasets for training purposes. Thus, we propose an unsupervised method to obtain the required labelled datasets for training the profile classifiers. Contrary to other author profiling approaches in the literature, we only make use of the users’ descriptions, which are usually part of the metadata posts. We exhaustively evaluated the proposed method under four different tasks for multidimensional author profiling along with state-of-the-art text classifiers. We achieved performances around 88% and 98% of F1 score for a gold standard and a silver standard datasets respectively. Additionally, we compare our results to other supervised approaches previously proposed for two of our tasks, getting very close performances despite using an unsupervised method. To the best of our knowledge, this is the first method designed to label user profiles in an unsupervised way for training profile classifiers with a similar performance to fully supervised ones.