Request Generator: g-REQ

 

Principle


The Generation of Requests consists of expanding a portion of text and relating it to all the terms present.

The expansion is done automatically, starting from information contained in a dedicated database called the "knowledge database" or KDB.

Proximity of the terms generated with the initial text (typically: a request by a user) is variable :

  • At 100%: it is considered that the generated term cannot be used in place of the original term in any way.
  • With a minimal value set it is permissible for the generated term to correspond to a word or an expression that may, semantically, be a long way away from the original term.

 

To generate a hypothesis the system builds a graph, i.e. a network of terms derived from those of the initial request, starting with linguistic information :

  • Phonetics: words which "sound" like the initial word
  • Morphological: words which contain (almost) the same letters, in (almost) the same logical order, or all of the words in the same way.

To create similar proposals for related terms, the hypothesis generation will reference the KDB which in turn has to contain the relevant information. To enrich the KDB the vocabulary derived from the process of hypothesis generation has to be inserted and this process can either be completely or partially automated via the use of complementary AMITM components: g-MIL, Marker Generator and g-EXT, Terminology Extractor.

The derived vocabulary consists of

  • Words
  • Expressions
  • Synonyms and acronyms
  • Close (related) requests.

 

The insertion into the KDB is done starting from the sources of information available and in addition to any dedicated capture process

The insertion process, and in particular the generation of "near requests", can be used to build and feed a log of requests.

  • within the application concerned
  • by the function dedicated to initialisation of the expansion.

 

An identifier can be associated to each request of the KDB in such a way that the search algorithm will provide support for the requests of users who are identified as sharing a similar vocabulary.

g-REQ provides automatic language recognition (and the encoding) of a word or a text, in English, French, German, Spanish, Italian, Portuguese and Dutch.

Applications

 

g-REQ can be used to :

Generate variations of a request

In an application such as a "search engine" and according to the level of language and interrogation used, it is possible to automatically generate various alternative suggestions or those which the user may not have thought of. It is also possible to offer interactive functions such as "test it with this" or "did you mean".

Propose Similar Requests

Sometimes called "Collaborative Filtering", this functionality makes similar search topics already submitted by others available to the user.

Manage a helpdesk and FAQ

The ability to access "already asked questions" and hence the corresponding answers allows fast and easy improvements in the effectiveness of call centre or helpdesk telephone support services.

Enrich categorisation and de-duplication

To produce a graphic understanding of the balanced assumptions of Noun Phrases derived from g-MIL makes it possible to increase the overall significance and broaden considerably the possible applications of this technology.