Many companies face an increase in data volume of unstructured text documents from customer feedback (surveys, customer support, chatlines, correspondence) or from company-specific documentation.
There is a need to quickly understand the content of such data and its main topics. User-tagging of large volume of text data entries is expensive, time-consuming and user-dependent.
A solution is to automatically assign tags to such datasets. Tags can be extracted directly from the text data or can represent text content at a semantic level.
Tags can have multiple uses:
- Re-structuring data collections to support classification of data;
- Information retrieval based on new and more relevant attributes;
- Finding documents linked through common tags.
TagTell is an interactive web application based on unsupervised key-phrase extraction methods and Natural Language Processing.
Three domain-independent, unsupervised approaches are implemented:
- Key-phrase extraction from text & various ranking mechanisms (term frequency, heuristics and graph-based co-occurrence);
- Similarity-based clustering of word embeddings;
- Using external resources for tag refinement and augmentation (Wordnet lexicon hierarchy).
The TagTell demonstrator provides an interactive platform for automatic tag extraction enabling:
- Tag extraction, ranking and visualization as a network graph;
- Tags updates (select, delete, replace tags with user-defined tags);
- Drill-down to find all documents with a certain tag;
- Download the data with added tags.
- Jayadeep Kumar Sasikumar, DIT
- Jimmy Doré , DIT
- Dr. Tamara Matthews, DIT
- Prof. Sarah Jane Delany, DIT