SentiAnswer is our Q&A system for open domains mounted on SentiLecto that extracts specific responses from thousands of texts
SentiAnswer first analyzes your question and extracts all the necessary linguistic information. Based on this, the most relevant documents are selected and analyzed to choose which specific snippet answer your query.
SentiAnswer can be adapted to any domain of knowledge with the simple change of its knowledge base, without the need for prior training, unlike other machine learning standalone algorithms. This entails a drastic reduction in the time invested to pick-up golden nuggets out of massive datasets of documents. SentiAnswer can be used as the NLP layer for virtual assistants (chatbots in call centers, for example), applications in LegalTech and the healthcare industry, FAQs about software documentation and knowledge discovery in general.
For an important legal publishing house in Argentina, we adapted SentiAnswer to be integrated into a search engine for specific questions about law cases. The idea was born from the observation that lawyers waste too many hours looking for cases to support their working hypotheses.
Civil Law, for example, is very different from Labor Law or Criminal Law. The law is unique but the interpretation of the judges is variable and that could be of value. They were facing the traditional problem of training and understanding natural languages the way native speakers that machine learning standalone algorithms usually encounter.
For example, when faced with questions like: ’Should the complementary annual salary be included in the calculation of a contract termination payment?’ We could isolate the operative parts of the relevant judgements (snippets) and answer the question (knowledge discovery), according to the knowledge base of Argentine jurisprudence: ’60% of the judges say NO and 40% say YES’ and then, we provided the specific fragments within the very long document conveying the judge’s position.
Now we can apply SentiAnswer on any other knowledge-base.