Social Sentiment Indices powered by X-Scores

SSIX Sentiment Analysis and Opinion Mining platform intends to monitor specific topics that are discussed in texts from several sources, including social networks, which are rich repositories of people’s opinions.

While the technology that is currently being developed is open and flexible to application to various domains (i.e. politics, sport, etc.), the data filtering component has currently been focused on some economic indicators that are relevant to our partners.

As discussed in a previous post, we are interested in people’s opinion about companies, products or even market segments. In any case, our interest is not only synchronic but also diachronic. We want to know, in fact, also how people’s opinion about such topics changes in relation to specific events.

In both cases, our task is to retrieve specific information about an object or class of objects (company, product etc.), investigating then how they vary in relation to specific events, such as for example the launch of a new product or a public announcement that the company has made.

In this perspective, it is clear the utility of monitoring social networks, where opinions constantly vary according to the upcoming events. In case of the product launch we would like to know what people think about a product before its launch, during its launch and after it. Which are their expectations? Are people interested? Did the interest turn into buying the product? Are they satisfied? Do they criticize something? What? And why? These are only few of the questions that SSIX can monitor.

Suppose someone finds a bug in a new phone just after its release and starts discussing about it on some public source. Clearly, being able to detect such discussion immediately has a huge economic impact for the releasing company. We want to know the people’s reaction on the short term, ideally in real time so as to immediately take due action. We want to know what triggered such reaction, who shares it, so that we can perform a root cause analysis and address the issue.

The release of a new product is just one example of the kind of “events” we are interested in. The SSIX platform intends to monitor public opinion also towards other kinds of events among which:

  • Company high level appointments
  • Merging between companies
  • Special wins, deals or customers that a company gains
  • Public announcements from a company like opening of a new hiring campaign, entering in the stock market, etc.

The consortium is currently developing technology capable of identifying and retrieving text relevant to the aforementioned categories.

A statistical classifier has been developed and trained on a manually annotated gold standard to label relevant text according to the event it describes. For the training we use a relatively small amount of data manually annotated with the activity type labels. Partially normalized text tokens, together with some semantic labels provided by a dedicated Named Entity Recognition component (e.g. company names, product names, etc.), are used as features.

From a first investigation, we found that this classifier obtains an F1 score of over 0.8. More precise features and more sophisticated approaches are now being investigated to improve this already good performance. At the same time, the technology is open and suitable to adaptation to other domains, as discussed above. Further domains of application can easily be defined in the future by training the classifier on different kinds of events, that characterize other domains, like politics, sport etc.

This blog post was written by SSIX partner Matteo Capelletti at Lionbridge.
For the latest update, like us on Facebook, follow us on Twitter and join us on LinkedIn.

Author :

Leave a Reply