Browse Prior Art Database

Natural language semantics in support of knowledge base searching and computer aided troubleshooting

IP.com Disclosure Number: IPCOM000246121D
Publication Date: 2016-May-10
Document File: 6 page(s) / 49K

Publishing Venue

The IP.com Prior Art Database

Abstract

We propose a system and method of exploiting natural language semantics in supports of knowledge base searching and computer-aided troubleshooting. With respect to state-of-the art virtual assistants, the proposed solution allows working with semi-structured troubleshooting knowledge bases, as they are collected by human assistants, with few or no knowledge re-engineering. Feeding the system with new knowledge, both factual and terminological, doesn't even require to stop the service. Moreover, by adopting a 'mixed-initiative' approach (the user and the system may inquire each other), the system leverages on both human and artificial understanding, taking the best of either side.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 25% of the total text.

Page 01 of 6

Natural language semantics in support of knowledge base searching and computer aided troubleshooting

Computer aided troubleshooting of problems involves a computer agent that interacts with users, the latter entering natural language text to explain their problems. The system must analyze the user utterances (text) and exploit a knowledge base of known problems, symptoms and causes annotated with textual descriptions. We propose a system and method exploiting natural language semantics in supports of knowledge base searching and computer aided troubleshooting. The system includes the following building blocks :


1. A Knowledge base (KB

KB

             KB) holding data according to a schema possibly derived from an ontology. As an example, a KB supporting the troubleshooting of software problems for a community of users may hold data related to problems for which known solutions exist. The schema of such KB may relate entities such as Problem and Symptom to Cause and Cause to Solution. One or both ends of such relations may associate multiple items , making it difficult for a human to make objective deductions from the available knowledge . Some KB entity classes have members that users may be able to refer to by verbal expressions. In the above example, these are problems and symptoms that users may experience. Enabling the KB semantic search amounts to attaching a textual description to each member of such classes .


2. A Natural Language Processor (NLP

NLP

                    NLP) pipeline processes user sentences (utterances) and transforms each sentence into a sequence of annotated tokens.

a. One key module in this pipeline - the POS tagger - assigns each token a tag identifying a corresponding part of speech (POS), like "NOUN", "ADJECTIVE", "VERB" etc. A lexicon of the language may provide additional information about words appearing in the analyzed sentences. For our purposes, it is critical that enough information be available in order to distinguish grammatical tokens (determiners, prepositions) from lexical tokens (nouns, verbs, adjectives), since the latter are the carriers of meaning.

b. One or more Named Entity Recognition (NER) modules must be able to detect one or more consecutive tokens as being the name of some entity. The type of NER modules to be included depend on the domain(s) of the dialog exchanged between the (human) users and the computer agent. As an example, a system designed to help users in troubleshooting software problems may critically depend on NER modules specialized for detecting (variants of) names of software products.


3. A Word Net

 Word Net representing all relevant relations between lemmas, such as synonym, hypernim and hyponim and identifying the meanings associated to each lemma. This Word Net does not need to include all lemmas of the underlying language, only words associated to textual descriptions present in the knowledge base and the additional words connected to the former set by any of the supported relations (closure...