Draft:Gavagai Explorer
From Trash Wiki
Gavagai Explorer is a text analytics tool developed by Gavagai used to process unstructured text data such as answers to open-ended questionnaires, comments, customer reviews, online mentions, customer feedback interactively by non-technical analysts. Gavagai Explorer is delivered as SaaS and is designed for professional but non-technical users, primarily to be used by market researchers, PR agencies, opinion analysts, academic researchers and in-house customer relationship management teams at large and small corporations and government agencies in a large number of sectors and fields in many linguistic and cultural areas.
[edit] Technical features
Gavagai Explorer clusters sentences from input texts in numerous languages automatically by topical themes and then scores them by tonality by a default 8-dimensional sentiment palette (general positive, general negative, love, hate, desire, fear, violence, skepticism) or by sentiment poles tailored for the task at hand. and allows the analyst to work with the quality of the clusters by interactively merging or discarding proposed clusters, and by inspecting the terms selected by the system as clustering criteria to refine the set. Unlike many other text clustering systems, Gavagai Explorer relies on term occurrence as a primary clustering criterion, instead of inferring a latent topic model. This is motivated by the desire of the designers to improve the transparency of the resulting analysis and the accessibility of the clustering criteria for human analysts: using words and multi-word terms explicitly mentioned in the text allows the analyst to immediately understand why e.g. a text is miscategorised and then adjust the cluster terms to reflect that error. Suggestions for terms to be included for improving cluster coverage are generated by consulting a back-end dictionary built by learning semantic relations from continuous monitoring of general language usage. Using latent (non-observable) variables instead may allow for greater precision but at a cost to editability and transparency.
[edit] Background
Gavagai Explorer builds on a back-end originally developed as a research project[3][4] by Jussi Karlgren and Magnus Sahlgren, founders of the text analytics company Gavagai. Gavagai was formed in 2007[5] to commercialise previous research by the founders on distributional semantics. Gavagai Explorer was originally developed to demonstrate the versatility of the approach to learning semantic relationships from text which is uniquely implemented as a form of the random indexing. The resulting learning lexicon was made available through a developer API but the interest for the demonstration tool was greater than the back end which today is the most important product for the company. Today Gavagai Explorer is used by market research companies and by academic researchers alike.
The back end technology has been used for media monitoring e.g. for public event outcome prediction such as predicting the winner of the Eurovision Song Contest, and for the analysis of political discourse. Recently (2018), the distributional semantics procedure has been applied to recordings of dolphin vocalisation. Gavagai has received numerous prizes and media attention for its technology approach and research background.
