CASE STUDY : INTELLIGENT SEARCH IN RESEARCH REPORTS
Our client, a leading global financial services firm, has many Research Analysts, who create a large number of research reports. These reports capture the analyst’s views on companies, sectors, commodities and regions. The reports have a lot of rich information, which is largely in unstructured or semi-structured form. These reports are consumed by research clients, who like to use the reports to answer specific questions that they may have. But the issue that the clients face is that it is very difficult to identify the right place in the reports, which would provide an appropriate answer to their questions. The firm wanted an intelligent search solution to resolve this problem. They wanted the search to return the most relevant results for their clients’ questions, by searching among the vast corpus of research articles that they have.
Wissen drew upon its Natural Language Processing (NLP) expertise to architect and implement an ingenious solution for this problem. It combined NLP techniques like Dependency Graph extraction, Word Embeddings and term frequency – inverse document frequency (tf-idf) to craft an elegant solution. The solution extracts triples from the unstructured text in the corpus and the questions. These triples enable contextual search, as they store syntactic and semantic information. The solution implements scoring and ranking algorithms, to bring up the right information from the reports, for a particular question. The solution gets information from different types of content in the reports, like paragraphs, tables (including tables embedded as images) etc. The other features of the solution are Rationale for the Result and Unsupervised Learning. It supports incremental load and is scalable while also supporting variations of questions.