The FITMAN “Unstructured and Social Data Analytics” (U-SDA) Specific Enabler (FITMAN-Anlzer) offers insights to unstructured data published in the web and to social data generated by users in social networks which are relevant to their specific topics of interest. Targeting the Digital Factory needs, FITMAN-Anlzer is based on state-of-the art, open-source technologies and is delivered as a web-based platform.
The FITMAN-Anlzer SE aims at extracting unstructured data from selected social media platforms and web resources and at turning such user-generated content to knowledge to be used for the benefit of any manufacturing stakeholder. It is responsible for:
- Acquiring information based on predefined keywords and / or accounts / pages,
- Filtering relevant information (i.e. removing keywords and URLs),
- Running opinion mining and trend analysis techniques (based on the SVM algorithm)
- Presenting emerging topics, sentiments and trends in an interactive way through intuitive visualizations (dashboards, tree maps, charts, galleries)
Conceptually, FITMAN-Anlzer is composed of different layers based on its various functionalities:
- The collection layer allows the SE to connect with the Twitter and Facebook open APIs to retrieve the desired data.
- The data level stores the collected data, as well as the user settings and the required keywords under which the analysis runs. It features two different repositories: (a) to store data as it comes to use them for the opinion mining analysis, and (b) to index them and process them in order to facilitate searching on the next level.
- The processing level allows for real-time search functionalities on the stored data, while it executes opinion mining algorithms to extract topics and conduct the sentiment analysis for them. In order to provide reliable results, it is required to train the opinion mining tool under a specific language, with a topic-specific training set.
- The view level allows the user to setup and manage the SE and is also responsible for visualizing trending topics and returning reports for user queries.
A high-level architectural overview of FITMAN-Anlzer is depicted in the following picture: