Computer Semantic Search of Inventive Solutions

Igor Boyko Ph.D
HPLab, Palo Alto, CA
igor_boyko@hpl.hp.com

 

ABSTRACT

The article covers problems of the Knowledge Base development and search automation of the inventive solutions by intellectual computer systems.

The following problems of the intellectual computer systems for inventive solution search are presented in the article:

The article presents design method of the inventive solutions semantic search on the basis of Universal semantic code (USC).

The article evolves of G.Altshuller's problem stating method in Algorithm of inventive problem solving and shows new possibilities of search of inventive solutions by the semantic tools.

 

1. INTRODUCTION

The problems solved by means of heuristic methods are considered invention.

For solving of such problems are developed the expert systems (ES). The Knowledge Base (KB) of any ES contains the facts and rules of their processing.

Inventive ES must help user to pass whole cycle through of creating the invention starting with the state of problem and finishing its solving. The knowledge in ES must be represented by determined way.

Practically, expert system is an intellectual system (IS), in which KB developing depends on chosen knowledge representation model.

Knowledge representation (KR) is oriented on supporting the possibility of computer solving of intellectual problems [2]. The choice of the KR model allows avoiding the complications at development of intellectual systems. However modern KR models cannot satisfy of the developer demands for KB of intellectual systems and particularly for inventive systems since they are built disregarding of natural language semantics [4].

About real intellectuality of the computer system possible to speak only if and when its KB is built on semantic basis. But semantic building of KB is unenforceable without using corresponding tools. One of possible tools is the Universal semantic code (USC) [6].

 

2. SEQUENCES OF ENGINEERNG OPERATIONS

The solution of inventive problem is arranged in the manner of patent for method, device or substance.

Where:

After analysis of patents for methods it was developed the order of verbs location in structures like this [7]:

'... verb 1 verb 2 verb 3 ...', which is equal in structures of type:

'... EO1 EO2 EO3 ...'.

It is used and more complex structure of type:

'... verb 1 verb 2 and verb 3 verb 4 ...', which is equal in structure of type:

'... EO 1 EO 2 and EO 3 EO 4 ...'.

Obviously each EO is located before or after some EO, i.e. all of EO can be simultaneously of two types:

  1. Previous;
  2. Subsequent.

The type of some EO depends on point of view it is considered.

 

3. PATENT ANALYSIS

Each invention has its problem of achievement of certain purposes. The problem state represents reached target situation in the manner of one or several EO.

Example. United States Patent 4 711 854.
Method of measuring moisture in a burnable absorber.
Goal of the invention -- measuring moisture in a burnable absorber.

DESCRIPTION OF THE INVENTION

Source of an inert carrier gas is provided that sends an inert carrier gas through line into enclosed chamber 1, which contains graphite crucible. A sample is placed within graphite crucible and crucible is heated through electrodes by means of an impulse electric current. Vapours from enclosed chamber 1 pass through conduit to enclosed chamber 2. Chamber 2 also contains a graphite crucible that is heated by impulse electricity through electrodes. Vapours leave chamber 2 through conduit and pass into hydrogen analyser.

In the first enclosed chamber, which contains the sample, means are provided to dissociate the water in the sample into hydrogen and oxygen. This is conventionally done by heating the sample in the presence of a dissociation catalyst such as graphite or platinum. The preferred dissociation catalyst is graphite because it can be formed into the shape of a crucible that can also be used to hold the sample, it is inexpensive, and a graphite crucible is conductive and therefore can be electrically heated by impulse heating.

While other means of heating the sample can also be used, electrical impulse heating is preferred because it heats the sample very rapidly so that any water in the sample is heated to its dissociation temperature while it is still in contact with the graphite.

We have found that a second enclosed chamber for dissociating water is necessary to achieve an accurate measurement of the hydrogen content of a sample. While the second enclosed chamber need not be the same as the first enclosed chamber, it is preferably identical for ease of construction and maintenance. The other portions of the apparatus, the source of inert carrier gas (e.g., argon, nitrogen, etc.), and the apparatus for analysing hydrogen content are conventional and are well known in the art.

>From given text the EO sequence of achievement of purposes 'measure' looks as the following sequence of verbs:

'send through heat dissociate pass through heat dissociate pass into measure'.

 

4. SHORT INTRODUCTION IN THE USC THEORY

The Universal Semantic Code is a semantic theory of Knowledge Representation [6]. With its help we can conduct semantic analysis of the target action and re-define it if necessary. The USC was developed as a semantic language of Knowledge representation and conversion. It has own algebra that uses the given set of variables.

The type of the USC algebra of formulas representation and conversion has been determined:

A = < M , * , - >

Where:

Principal unit of Knowledge representation of the USC is a complex formula that represents of variable interaction. Semantic relations in complex formulas are represented in the following way:

((X*Y)*Z)*(Z*(Y*Z)) - X by means of Y acts on Z and as a result Z keeps Y in itself
(in Z).

We would like to draw attention to the fact that the left part in this type of formulas causes potential action in right part. Thus complex formula reflects the following situation:

Stimulus Reaction.

Each complex USC formula has one and only one meaning and each conversion of USC formula into another one has one and only one meaning conversion. The rules according to which meaning conversion is done are contained in the USC algebra axioms.

USC formulas represent actions that are met with in any field of engineering. The USC allows structuring engineering actions database, turning it into Knowledge Base and enclosing semantics in representation formalisms.

The USC-classifier contains engineering actions list and USC formulas classified in a certain way. In the USC-classifier all engineering actions are analogues. Each main engineering action has own USC formula that determines this engineering action and verbal interpretation demonstrating the action at issue. Each action-analogue has own main action.

Fig. 1.1 shows the USC- classifier structure.


Fig.1.1

Practically, within the limits of the given classifier the non-traditional system of engineering actions-analogues is built according to the function they perform but not according to their achievement technology.

In computer intellectual systems for description of problem conditions necessary to use strictly structured of KR units.

USC formulas have single sense, strictly given length and limited set variables filled by specific dates. Involuntary sequences of USC formulas define the problem solving variants.

In the USC verbs classifier are presented meeting in technology verbs and are has given own interpretations in the USC formulas manner. Since each verb expresses the relation appearing between subject and object of action in the USC classifier is presented structures of proposed relations in evident type. Such structures are knowledge representation units.

In USC classifier all verbs are main or inherit. Every USC formula corresponds to main verb and each USC formula has own verbal interpretation. Each inherit verb corresponds to the main verb. Practically main verb is the most abstract presentation for a certain EO.

For example, the verb 'heat' is the main and is presented by USC formula:

heat ((X*Y)*Z)*((Z*Z)*Y) -- 'X' by means of 'Y' heat 'Z'.

The verb 'measure out' is related with the verb 'measure' that is the main and is presented by USC formula:

measure ((X*Y)*X)*((X*Y)*Z) -- 'X' by means of 'Y' measure 'Z'.

Practically, within the framework of given verbs classifier is built off-center actions-analogues classifier or in other words EO classifier.

 

5. USC THEORY APPLICATION

 

Given classifier was applied for analyzing of patents.

Lets we will find into the USC classifier all verbs that were got in EO sequence of the patent.
The verb 'send through is related with the verb 'orient' that is the main. The verb 'heat' is the main.
The verb 'dissociate' is related with the verb 'decompose' that is the main.
The verb 'pass through' is related with the verb 'conduct' that is the main.
The verb 'pass into' is related with the verb 'embed' that is the main.
The verb 'measure' is the main.

The final sequence of abstract EO for considered patent is following:

orient heat decompose conduct heat decompose embed measure'.

Each EO has own USC formula and verbal interpretation. Now we will fill of formula variables for considered patent.

'orient' ((X*Y)*Z)* ((Z*Z)*Z) ' X by means of Y orient Z'
X = source;
Y = line;
Z = inert gas.

'heat' ((X*Y)*Z)* ((Z*Z)*Y) 'X by means of Y heat Z'
X = electrodes;
Y = impulse electric current;
Z = sample.

'decompose' ((X*Y)*Z)* ((Z*Y)*Z) 'X by means of Y decompose Z'
X = graphite crucible;
Y = impulse electric current;
Z = sample.

'conduct' ((X*Y)*Z)* ((Z*X)*Y) 'X by means of Y conduct Z'
X = conduit 1;
Y = internal surface;
Z = vapours.

'heat' ((X*Y)*Z)* ((Z*Z)*Y)) 'X by means of Y heat Z'
X = electrodes;
Y = impulse electric current;
Z = vapours.

'decompose' ((X*Y)*Z)* ((Z*Y)*Z) 'X by means of Y decompose Z'
X = graphite crucible;
Y = impulse electric current;
Z = vapours.

'embed' ((X*Y)*Z)* ((Z*(W*Z) 'X by means of Y embed Z into W'
X = conduit 2;
Y = internal surface;
Z = vapors;
W = analyzer.

'measure' ((X*Y)*X)* ((X*Y)*Z) 'X by means of Y measure Z'
X = user;
Y = analyzer;
Z = hydrogen.

The representation of the whole process by means of filled by specific dates USC formulas allow to specify the subjects, instruments and objects for each EO.

 

6. CONSECUTIONS OF ENGINEERING OPERATIONS

Lets consider how reach the target situation that is determined by the verb 'measure by means of EO sequences. For example, from the developed base of EO sequences the some part of the previous verbs tree for EO 'measure is following:

There is the previous verb 'embed.
Now on the left of verb 'embed we will add previous him verbs. As a result we get the following verbs tree:

The following step, add the new verbs at the left part of the tree to verb 'decompose.

Further we form the tree with the left side to verb 'heat.

We continue to build the tree to verb 'conduct.

To each verb from extreme left part possible to add previous verbs list.

Obviously that for performing the operation is presented by verb 'measure possible to use the many ways each of which presents own verbs sequence, i.e. the sequence of EO.

Now instead of specific verbs we will substitute corresponding USC formulas. As a result we get the following USC formulas tree:

The sequences defining the order of using USC formulas (or EO) we will name the semantic units sequences (SUS) since each USC formula is an elementary KR semantic unit.

Practically, got USC formulas tree we can consider as a graph of which top are presented by USC formulas.

The target top is assigned the target verb.

The full sequence USC formulas for considered patent is following:

((Z*Z)*Z) ((Z*Z)*Y) ((Z*Y)*Z) ((Z*X)*Y) ((Z*Z)*Y) ((Z*Y)*Z) ((Z*(W*Z)
((X*Y)*Z)

You can see that if we have rules of conversion one USC formulas in others in the manner of USC algebra [6] we can calculate the ways of inventive problems solving. Then substituting instead of USC formulas some specific dates in the manner of verbs it is possible to examine the variants of solving of stated problem.

 

7. INTELLECTUAL PROBLEMS TYPES

At present solved by person intellectual problems possible to divide into two main types:

  1. Integrated-dispatching problems
  2. Inventive problems

Integrated-dispatching problems are problems on planning the displacement, keeping, creation and deleting the different types of resources both physical and information.

The inventive problems (IP) appear when inventor develops of principal new or improves already available products and technologies. As a rule IP solving is reached at permit of different contradictions types [1].

Inventor begins solving of IP after stating initial and target situations (IS and TS). I.e. transition algorithm (TA) from IS to TS is a way of IP solving. Both IS and TS for the correct keeping of formulations must be presented by single verbs.

Imagine you have KB of transition algorithms in the computer system and the KB is built on the base of USC formula conversions. Comparing each of formulated verbs with available KB of transition algorithms possible get several alternative TA from IS to TS. I.e. inventive problem is reduced to integrated-dispatching.

Since TA already exists it need only consecutively to pass through its steps for specifying of objects instead of variable.

For inventive problem solving it is necessary to present only TS. Moving from TS to previous EO inventor solves on what previous step he will stop, i.e. defining this step as preferred IS. Besides, in process of retrospective motion possible to solve some contradiction that defines the appearance of inventive problem.

Practically, set of SUS presents the graph with limited by set of elements (the verbs) strictly bound between itself and describing inventive problems world, as well as integrated-dispatching problems. So amount of both types of problems is limited by the amount (though and enough extensive) of the SUS. Appearance of new substances and materials does not enlarge this amount but will only initiate of using of SUS not used earlier.

 

8. CONCLUSIONS

  1. Using the more abstract verbs by means of USC classifier of EO allows reducing the psychological inertia obtruded by specific terms of some application domain.
  2. Distraction from specific descriptions by means of abstract descriptions of processes allows finding of inventive problem solutions by analogy. Well-known inventive problems solving by means of analogies practically is a main tool of searching for formulated purpose [1; 5].
  3. Filling of USC formulas by variables specifies the subjects and objects considered EO and can force the inventor to move to physical micro-level. This initiates deeper understanding of considered situations.

9. REFERENCES

  1. Altshuller G.S. Creativity as an Exact Science. Gordon and Breach Science Publisher, 1984.
  2. Ueno H., Koyama T., Matsuby M. Knowledge Representation and Using. Isidzuka. Omsya. Tokyo, 1987.
  3. Martynov V.V, Boyko I.M., Gyminski A.P. Semantic Coding and Invention Problem Solving. // TIPS journal. Obninsk, 1991, N2.1. (in Russian)
  4. Martynov V.V, Boyko I.M., Gyminski A.P. Knowledge Bases Construction of Systems for Solving Intellectual Problems. // Controlling Systems and Machines. // Kiev, 1992, N5/6.
  5. Altov H. And Suddenly the Inventor Appeared. Published by Technical Innovation Center. MA, USA, 1994.
  6. Martynov V.V. Universal Semantic Code: USC -5. MSLU Pre-print N4. Minsk, 1995.
  7. Boyko I.M. Semantic Search on the Stage of the Structure Designing in the CAD. PhD thesis. Institute of Technical Cybernetics of Academy of Science of Belarus. Minsk, 1995. (in Russian).

 

About the Author

Igor M. Boyko
Date of birth: 22 may 1960

1981-82 He learned TRIZ in Belarussian State University of Informatics and Radioelectronics (former Minsk Radio-engineering Institute).

1982 He finished the University and worked as a computer adjuster for a long time.

1987 In Academy of Science of Belarus he began the development of computer application of natural language semantics in TRIZ. His main field was the Knowledge Representation in Artificial Intelligence.

1988 It was the personal acquaintance with Altshuller G.S. in Leningrad before the TRIZ conference in Petrozavodsk.

From 1988 Igor Boyko actively teaches in Belarus TRIZ schools. He develops own semantic methods for application in TRIZ and tests the methods with pupils.

Igor Boyko took part in Altshullers TRIZ seminars and conferences that took place in former USSR. He kept the correspondence with Altshuller G.S. about TRIZ questions.

1995 He defends the PhD thesis Semantic Search on the Stage of the Structure Designing in the CAD where he analysis semantics as tool of inventive problem solving in the CAD. He compares traditional invention methods, tools of TRIZ and his new approach of semantic application for inventive problem solving.

1995-2000 He took part in development of different directions of Invention Machine project and continued development of computer semantics methods for inventors.

At the time Igor Boyko works in HP Laboratory where develops new approaches of natural language processing and understanding for computer users.