Name of the participant: Kathrin Eichler
Description of the IT-research project: Since human language allows us to formulate a request in many different ways, it is not sufficient to compare texts on a purely word level, i.e. based on the number of words that match in the texts. An internal analysis on a large amount of data from German customer inquiries has shown that only for about 12% of the customer inquiries the matching result could be found by a search that compares texts on a pure word level.
The aim of the project is the development of an inference-based search function which, unlike conventional search functions, analyses and compares the query text and the text data already entered in the system on a content level. This should be able to be used to find suitable problem cases that have already been recorded in the system for a customer enquiry received by the customer support department.
Methods that can be used to determine whether one statement follows on from another are being developed within the still relatively new research area of RTE (Recognizing Textual Entailment). Existing RTE systems have so far only been used in research environments and evaluated on “clean” texts such as newspaper articles. Within the project, it will now be investigated how RTE can be used to analyse customer queries on a semantic level and to evaluate how well a query fits to a problem case recorded in the system.
The goal of the project is to develop a prototypical system that is able to automatically assign an incoming customer inquiry to an existing problem category without the need to match the words used in the inquiry and problem description. For this purpose, linguistic tools and knowledge sources will be used and it will be investigated to what extent RTE components can contribute to the improvement of the system. The system should take into account the specific requirements of the use case.
Software Campus partners: DFKI, Software AG
Implementation period:01.03.2013 – 28.02.2015
