FactMind - automatic intelligent answering of any question

with regard to any text

Project homepage:   www.PreciseHelp.com (Free Internet FAQ search engine)

Project homepage:   www.FactMind.com (identical to these pages)

Answers any question with regard to any text on any domain at any time

instantly and automatically!

High-Tech adjusting itself to human requirements and not the other way around!

Presumably the most intelligent search engine & knowledge management system for information deposited in texts / web pages, since no stop word elimination takes place, but the meaning of these stop words - e.g. prepositions - is being considered. This allow the up to now largest possible automation in the area of knowledge management and question answering.

What is FactMind?

FactMind is easy to handle and widely applicable - which becomes possible by harnessing the latest and most advanced technologies: FactMind answers textual inquiries in natural language by interpreting the meaning of queries and document collections and finally providing relevant passages. This type of application can be compared to the classical Internet search engines or office software assistants - but employs a brand-new approach. This technology speeds up and simplifies your daily work and improves the competitiveness of your company drastically due to lower cost structures.

Which are the advantages of FactMind?

The main advantages of this innovative technology are:

  • usage, maintenance, and the deposited information can be in natural language
  • reduction & resolution of ambiguities
  • search over tags and meta tags (for files in the formats HTML, SGML, XML)
  • search dependent on text formatting (bold, italic, underlined), positioning in headlines, titles, foot notes or other tags in the context of HTML, SGML or XML files
  • no time required for learning specific query languages
  • the information contained in documents like refinements / variations of queries can be displayed hierarchically and be used for guiding / helping the user
  • no stop word elimination (like prepositions, conjunctions) that would mean a loss of information
  • consideration of relations based on meaning like more general terms, more specific terms, synonyms, antonyms, elements of, parts of
  • consideration and weighting based on linguistic relations covering nested phrases as well as entire paragraphs; most systems do just use the distance between key words to calculate a score
  • consideration of the type of agreement between query and relevant document positions in terms of linguistic roles like subject, predicate, object, and other relations (e.g. which words are described in more detail by means of paraphrases, adjectives, adverbs).

In how far is FactMind more flexible regarding questions and deposited knowledge?

FactMind is - in contrast to today’s common applications - equipped with a much higher level of text understanding ability.

For your questions you may use:

  • classical keywords,
  • free formulations,
  • creative neologisms / composed words,
  • circumscriptions for a technical term, that may be difficult to remember,
  • synonym terms, generic- or subordinate terms,
  • prepositional or adverbial relations and
  • one or several complex sentences

FactMind understands you. This means concretely: You pose questions just like you would when in conversation.

What can FactMind be applied to?

FactMind is very widely applicable – literally in all fields related to texts or knowledge and queries. Some examples:

  • Support hotlines / call centers / helpdesk systems / online customer support: FactMind tries to answer inquiries arriving through web forms, email, fax, and, upcoming, by phone, automatically in first level support. Due to call avoidance alone, customers avoiding phone calls in favor of textual media like the Internet, about 50% of the costs can be saved. FactMind recognizes whether it has found a relevant answer and if necessary passes requests on to the first or second level support.
  • Internal knowledge database: Coworkers make use of the entire know-how of an organization, 24 hours a day, quickly and effortlessly: This way, the coaching of new colleagues is shortened and mistakes arising from insufficient information can be avoided. Documents, which have been created within the frame of the DIN/ISO 9000 certification process, can therefore be used directly.
  • Didactical materials, instructional books, CD-ROM courses: Questions of students can be answered automatically. So the main disadvantage of autodidactic learning is mostly compensated.
  • Open FAQ database (frequently asked questions and their answers) for the general public or associate partners: With FactMind each group gets the quick and exact information it needs. Also solutions proposed by third parties, in chat forums or newsgroups, can be automatically integrated.
  • Help assistants, selling assistants, avatars, chat robots applied to complex applications, help on the filling of forms, products sold in online shops - example: "Which products would supplement those in the shopping basket in such a way that they are recommendable for a well-balanced lunch, low in calories and are at the same time good for more than two weeks?"
  • City, company or product information systems: Instead of working through menus and comparing different offers, everything can be done with a single question. Example: "Which hotels have vacancies in double rooms for which they charge up to 100 dollars within one kilometer distance of the railway station?"
  • Natural language data base querying / consulting lexicons & encyclopaedias - example: "Which are the committees of the European Union?"
  • Improved information retrieval systems / Internet search engines
  • Recommender systems / matching of corresponding or similar descriptions - like reusable software components, partner search, substitute products
  • Product documentations on CD-ROM / DVD or in the Internet: Technicians, work shops, users, system integrators get their questions directly answered without complicated searching or browsing in inferior figures.
  • Advisory or referrer services in such areas as medicine, law, and taxation.

Integration of the helpdesk module

 

 

FactMind is continually improved by Compris Intelligence GmbH in cooperation with Fraunhofer society and by implementing the latest scientific insights in computational linguistics. It is therefore able to meet with the ever changing requirements of today's world.

FactMind as a concert of three technologies

FactMind can import knowledge from arbitrary texts that can each be stored in various textual file formats. The whole process starts by depositing the meaning of texts in a sophisticated knowledge representation formalism. Questions are brought into this knowledge representation in real time. These exact representations of text meaning - mostly without ambiguities - will then be matched against each other. In a first step FactMind identifies the matching key concepts with the speed known from Internet search engines. The second step features the subtle linguistic evaluation of the candidate matches from the first step. For this reason this approach also works as a meta search engine in which classical search engines carry out the first step.

Linguistic parsing technology

By employing the modern HPSG (Head-Driven phrase Structure Grammar) language technology of the CSLI (Center for the Study of Language and Information; Stanford University, USA) vocabulary can be covered better. This technology is able to learn and represent any semantic relations.

 

A general English parser is used, which constructs a parse tree, resolves ambiguities, performs morphological reduction to basic forms and reduces ambiguities. Furthermore, the semantic roles of the words (subject, predicate, direct/indirect object) are determined.

To make it more readable, the parser output is often represented as a tree. Out of that tree, the dependencies found between the sentence elements can easily be read.

 

Pattern matching on text and tags

SGML, HTML, XML - all have tags in common that contain further information on content or formatting. In addition to the classical search in content it is also possible to look for particular combinations of tags and content: For example it is possible to search

  • for the word "accident/accidents" mentioned in a headline or
  • for specially marked information in laws and not in comments or
  • for product names that appear as bold or italic type faces.

Another highlight is the possibility to specify that key words or key concepts in headines or in highlighted form should have a higher weight. Smaller weights are typically assigned to larger accumulations of key words in hardly readable font sizes at the end of web pages. Through a particular marking in the knowledge base it can be guaranteed that mandatory legal disclaimers will appear in any case in the output.

 

Statistical weighting & scoring

Humans put important statements at the beginning of a sentence, putting less important information into subordinate clauses. It does usually make a difference if something is mentioned as subject, predicate or object. Frequent words, in particular filler words are often less important. Prepositions (before, while, above) as well as conjunctions (with, because) are usually important and have their special meanings. Even the classical search engines take the number of appearances of key words into consideration. Exactly these factors are combined statistically in FactMind following an elaborate system. This technology also decides - depending on the context - which of several imaginable meanings is present. The results are presented in the order of their ranking. Due to this approach the system automatically recognizes when it has not found an appropriate answer and can forward the query to humans. You are likely to find that there is no appropriate answer in the underlying knowledge base in such cases.

Which improvements are in progress?

  • The identification of linguistic references ("anaphora") such as "he", "these", "the former", "their", "second", etc.
  • Exact consideration of temporal and logical relations, in particular causal chains, reasons/symptoms, chronological processes
  • Better recognition of descriptive wordings
  • Versions in further languages

Which versions can be delivered?

Versions are available for MS-Windows as standalone systems and software development kits (SDK) including compiled libraries for linking them to applications. For Unix, in particular Linux, a web server version is additionally available. Currently, English and German releases are shipped.


 

Address

Compris Intelligence GmbH
Rheingönheimer Str. 79
67065 Ludwigshafen am Rhein
Germany
phone: (+49) 0700-COMPRISTel (0700-26677478)
fax: (+49) 0700-COMPRISFax (0700-26677473)

Internet: www.FactMind.com
E-Mail:products@compris.com



Information & Questions: products@compris.com


www.compris.com  | FactMind  | Contact/Map  | About Compris Intelligence GmbH