Neural network for computer-aided knowledge management

• US 7,092,857 B1
• Filed: 05/24/2000
• Issued: 08/15/2006
• Est. Priority Date: 05/24/1999
• Status: Expired due to Fees

• Current Assignee: Ipcentury AG
• Original Assignee: Ipcentury AG
• Inventors: Steiner, Ralf
• Primary Examiner(s): PALADINI, ALBERT WILLIAM
• Application Number:
  US09/979,684
• Time in Patent Office: 2,274 Days
• Field of Search: 703/22, 703/2, 706/12, 706/16, 706/45, 706/50, 706/61
• US Class Current: 703/2
• CPC Class Codes: G06N 3/042 Knowledge-based neural networks; Logical representations of neural networks

Abstract

The invention relates to a method and a neural network for computer-assisted knowledge management, based on a neural network (1) that is formed by a computer in its memory location. The invention method and neural network are especially for use in a decentralized, computer-assisted patent system that can be used via the Internet system, in the broad sense. The neural network (1) forms a system of artificial intelligence (KI), covering a fundamental knowledge base in the form of computer-readable texts. The neural network (1) consists of elements (2) that are associated with each other and weighted in relation to each other so that the sets of knowledge available can be managed and analyzed in relation to each other by computer means.

Claims

1. A neural network for computer-aided knowledge management, with the neural network being composed of elements which are related to one another and are weighted, and which network is stored computationally as an associative data structure dynamically in the memory area of a computer, and the individual elements are allocated a significance content, characterized

in that each element is allocated a definition of a development as the significance content, or directly contains this,

in that this definition forms a subset in the form of an interaction pair which, optionally in the form<

quantity|quality>

via its verbal definition, defines the development and hence accurately defines the knowledge set, which maps the quantity onto the quality, and the quality onto the quantity, and

in that the elements form a Hilbert space.

2. The neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that an operator is implemented for calculating a scalar product between each of the two elements, and is optionally in the form of the Euclidean for the n-dimensional space,

in that an associated NOT element with a negative weighting for the element can be generated by the neural network for each element, which NOT element accurately denotes the knowledge set excluded from the knowledge set of the associated element, and is optionally generated as a pair with the element,

in that a new entry of an element is implemented in the neural network as an addition, and a weighting of a reference is implemented as a multiplication in Hilbert space, and

in that a mass can be calculated for each element by the neural network, and is optionally stored with it.

3. The neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that the element has one, and only one, associated coordinate vector, or directly contains this vector, which is optionally calculated via recursions in the neural network,

in that the coordinate vector is optionally in the form of a dynamically concatenated list and optionally includes only the necessary coordinates organized with and on the basis of their time index,

in that, when entered, each element optionally includes a unique time index and a unique user index,

in that encryption, identification of the authorized user and authentication of the inputs are optionally ensured by implementing cryptological methods, with the authenticity of the definition and of the references optionally being ensured by means of the time index and the user index, and the definition and the references optionally being encrypted,

in that the various types of list elements are optionally associated with the calculation time index of their last calculation via a temporary data area, with this calculation time index optionally being compared with the current calculation time index before carrying out the calculation, and

in that each element optionally has a cross-reference to the associated entry in a content database, which is used as the basis for generating the knowledge.

4. The neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that the neural network is a component of a server,

in that it is connected to a user via a dialog system, optionally via the internet and optionally with the inclusion of a secure e-commerce system,

in that inputs for the neural network are optionally entered via a chronological stream,

in that separate decentralized neural networks are optionally connected to the central neural network via a communication link and are optionally half-mirrored, and

in that the dialog system optionally has an intuitive interface which, as an input for the neural network, allows a definition of the interaction pair, from a selection of references selected via switches, and optionally allows a short title and a short description and, as an output, allows the graphical display of a local environment of directly connected and/or unconnected elements in the associative data structure, and has further tools.

5. The neural network for computer-aided knowledge management as claimed in claim 1 characterized

in that the neural network has specifically matched associated systems for applications in patent matters, macroeconomy, market analysis or gene analysis,

in that, specifically for use in patent matters, a laid-open period of grace is optionally provided for the output via an appropriate time filter, the knowledge of a protective right is optionally identified as a copy with final references which cannot be varied, with definitions and with a particular flag, a computationally defined substantial novelty with regard to the totality of the significant references and a computationally defined substantial inventive step in comparison to the mass of data can optionally be calculated with respect to such protective rights knowledge, relating to a given time index, in comparison with the associative data structure, and a cross-reference is optionally included to the associated official register with the legal situation.

6. The neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that each of the individual elements has an associated supplementary verbal text field which briefly explains the significance content of the knowledge set for the above definition, or represented in abbreviated key word form, with the individual elements optionally having both explanatory and abbreviating associated text fields in the form of a short title and short description, and

in that each part of the interaction pair is optionally subdivided into a short version and a definition and has an associated text field.

7. The neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that the neural network has associated means for telecopying and/or for long-distance transmission of text data, and

in that means are optionally provided for supporting electronic signatures.

12. Use of the neural network for computer-aided knowledge management as claimed in claim 1, characterized

in that, for patent matters, it is used as an expert system for substantial decisions, for administration or for analyses, with the interaction pair being formed between solutions and tasks or means and effects in the technical field, or

in that it is used in the macroeconomy as an expert system for financial decisions, for administration or for analyses, with the interaction pair being formed between individual economies which can be separated financially and activity fields, or

in that it is used in marketing as an expert system for marketing decisions, for administration or for analyses, with the interaction pair being formed between products and customer requirements, or

in that it is used in gene technology as an expert system for DNA sequencing, for administration or for analyses, with the interaction pair being formed between gene sequences and biofunctions, or

in that it is formed in the innovation requirement for protective rights application by innovative developments at protective rights receiving points.

8. A method for a neural network for computer-aided knowledge management, characterized in that

an expanded input to the neural network is entered in one step, having optionally being converted in advance to a stream,

linking and incremental expansion of the associative data structure of the neural network relating to this input being carried out in a further step by access to a linear memory area of the computer, the use and arithmetic of computer-specific pointers to parts of this memory area and a stack which is sufficiently large for recursions over the entire associative data structure,

the weight of the elements being updated and/or calculated in a further step, with the new entry of an element corresponding to an addition and the weighting of a reference corresponding to a multiplication using the implemented operators for calculating the metrics, norm and scalar product in Hilbert space which is preferably in the form of the Euclidean for the n-dimensional space,

convergence of the associative data structure of the neural network for low redundancy is carried out statistically in an optionally large number of the preceding steps,

administration of and/or analysis between individual elements is carried out in a further step, using the operators for calculating the metrics, norm and scalar product in Hilbert space, via the calculation of masses by the neural network, and

the results of the administration or analysis are output in a final step.

9. The method for a neural network for computer-aided knowledge management as claimed in claim 8, characterized

in that all the elements are stored dynamically, and are administered dynamically via a list concatenated via pointers,

in that each element has a uniquely associated NOT element, which optionally has a pointer to the same interaction pair and is optionally generated with the generation of the element, with the NOT element optionally differing from the associated element in the LSB of the time index,

in that the generation of the references themselves is optionally carried out via interval interleaving to determine that memory area of the reference element which is associated with the time index in the specific associative data structure, and

in that this is optionally concatenated at one end or alternately via pointers, which are optionally administered via dynamic reference lists,

in that the references are themselves optionally provided with dynamically stored, optionally user-specific weightings, which are optionally formed via the LSB of the unique user index and which are optionally administered via dynamically concatenated weighting lists,

in that the operators for calculating the metrics, norm and scalar product in Hilbert space from the neural network are applied to recursively determined coordinates and/or to coordinate vectors of the elements, and the analyses are optionally based on vector calculation methods, and

in that a classification system is optionally generated by the neural network from the most significant references of the associative data structure and from the associated elements.

10. The method for a neural network for computer-aided knowledge management as claimed in claim 8, characterized

in that, even while the definition of the interaction pair is being entered, a selection of elements is output from the neural network via the intuitive interface, which elements contain similar terms to those in the definition, with a reference (which is connected by the user) to an element optionally being used as a training input by the neural network,

in that new quantities, which define the described knowledge set, or qualities are optionally defined (in terms of the terminology) by the user on the basis of the quantities and qualities, respectively, which are already known by means of stated references, with AND, OR, NOT operators optionally being used, which are optionally transformed by the neural network, by means of analysis of the syntax and semantics, to mathematically logical terms of terminology and being solved recursively by computation via the reference elements, with the neural network in each case optionally using this to generate a respective Thesaurus for quantities and qualities, which contains these terms on a hierarchically structured basis and weighted in accordance with the weight of the elements which use them,

in that the user optionally adds to each element a specific development, which is contained in the described knowledge set, as a brief description and a brief title, in which case the brief descriptions and definitions can optionally be searched for terms in the normal manner on a text-oriented basis and/or via linguistically worthwhile links in the context, with the neural network optionally for this purpose generating in each case one right-ordered and left-ordered indexed list of the terms with a cross-reference to the time index of the element containing it, or a pointer to it being generated, and

in that the stream which is used for the input is optionally generated automatically during preprocessing from text-based contents of content databases, in that a computer-aided, automatic, compressed projection from the context of the content is produced into the format of the stream, with the search options and the Thesaurus for the knowledge base of the neural network optionally being used automatically during the preprocessing.

11. The method for a neural network for computer-aided knowledge management as claimed in claim 8, with the innovation being represented as a specific development relating to a computer-aided knowledge management system via verbal definitions and references to other already known developments, characterized in that

the innovation is focused successively by the user, with regard to high innovation potential, in an optional first method step relating to the novelty and inventive step calculated by the knowledge management system and output to the user,

in a further method step, the user enters the details of the identity and address of the user and all the text fields,

in a further method step, the development is entered into the knowledge management system as an innovation, with this development optionally not being accessible during a publication period of grace by being securely encrypted on a user-specific basis,

in an optional further method step, a provisional protective rights application is generated with computer assistance using standard phrases, and

in an optional final method step, these are transferred without delay to a protective rights receiving point, with this application optionally being signed with an electronic signature.