CN104331523A - Conceptual object model-based question searching method - Google Patents

Abstract

The invention discloses a conceptual object model-based question searching method which sequentially comprises the following steps: building a domain concept object model in accordance with a conceptual object model knowledge representation method according to the characteristic of a research definition field question; inputting an enter text by a user and determining the enter text; defining a similarity calculation formula of the question based on a conceptual object model; and performing question searching based on the similarity calculation formula of the question. According to the conceptual object model-based question searching method, the semantic analysis can be performed, the efficiency is improved, and the time complexity is reduced.

Description

A kind of question sentence search method of concept based object model

Technical field

The present invention relates to automatically request-answering system research field, be more particularly to Similarity Measure and the question sentence retrieval research of question sentence in automatically request-answering system.

Background technology

The Similarity Measure difficulty of Chinese sentence is very large, and classic method mainly comprises based on TF-IDF method and the algorithm based on semanteme.TF-IDF method based on vector space model is mainly used in large-scale file retrieval, and only have when the word that sentence comprises is abundant, the method just has reasonable effect.Question sentence due to restriction field is short and small and the vocabulary quantity comprised is few, so effect is not good enough.TF-IDF method only considered word statistical property within a context, does not analyze on semantic level, well can not embody the characteristic of Chinese, have certain limitation.Algorithm based on semanteme: due to the restriction of semantic data bank, efficiency of algorithm is low, the real-time of question answering system is deteriorated, Similarity Measure based on semanteme needs the support of a certain semantic knowledge dictionary, and the mainly world knowledge comprised in knowledge dictionary, and for a certain professional domain knowledge package contain less, lack the support to professional domain keyword, less effective calculated to the Question sentence parsing in field.

Therefore the present invention researchs and analyses by calculating Problems existing to the relevant feature of domain knowledge and current statement similarity, setting up on the basis of conceptual object model in conjunction with certain semantic analysis, propose the Question sentence parsing computing method of concept based object model, the method can be carried out the analysis of field concept Object Semanteme and improve efficiency of algorithm, reduces Algorithms T-cbmplexity.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, provide one to carry out simple semantic analysis and to improve efficiency of algorithm, reduce the sentence retrieval method of the concept based object model of Algorithms T-cbmplexity, in turn include the following steps:

Step 1: according to the feature of research restriction field question sentence, according to conceptual object model knowledge representation method, set up field concept object model;

Step 2: user input text question sentence is also determined;

Step 3: character crossfire process, extracts concept, object and the attribute in question sentence, utilizes the theme of concept or object factory question sentence, utilize the attribute description question sentence focus of concept or object, question sentence is expressed as conceptual object model representation form;

Step 4: the Similarity Measure mode of concept based object model definition question sentence, be specially and Question sentence parsing calculating is divided into four parts, Similarity Measure and remainder Similarity Measure between Similarity Measure, relation between Similarity Measure, attribute between object, last weighted calculation obtains the Similarity Measure mode of question sentence;

Step 5: the Similarity Measure mode based on question sentence carries out question sentence retrieval.

Preferably, described according to conceptual object model knowledge representation method, set up field concept object model, concrete steps are: therefrom extract field concept by analysis field question sentence feature, domain object, concept attribute and object properties, and the relation between them, obtain field concept set, domain object set, community set and set of relationship, the representation of the contextual definition relation by analysis in set of relationship, relation wherein between concept and attribute, inheritance between concept and concept, relation between concept and object individually represents, finally set up field concept object model, specific as follows:

Step 1.1: extract field concept, set up field concept set Concepts={C ₁, C ₂... C _n;

Step 1.2: extract domain object, set up domain object set Objects={O ₁, O ₂..., O _m;

Step 1.3: extract concept attribute and object properties, set up community set Attributes={A ₁, A ₂..., A _k;

Step 1.4: extract field concept, domain object, relation between concept attribute and object properties, opening relationships set Relations={R ₁, R ₂..., R _l;

Step 1.5: the representation of defined notion, the relation between attribute and object, the relation between concept and attribute, the inheritance between concept and concept, the relation between concept and object and other relations, sets up field concept object model.

Preferably, the concrete steps of described step 3 are as follows:

Step 3.1: utilize Forward Maximum Method method, finds out all spectra object in user's question sentence according to domain object storehouse, if question sentence is only containing a domain object, goes to step 3.2, otherwise goes to step 3.3;

Step 3.2: for domain object, according to field concept object model, obtain the attribute of its concept as object properties, the object properties utilizing Forward Maximum Method method to find out user's question sentence to comprise, if do not find object properties in question sentence, then provide a default property, the conceptual object finally forming inquiry question sentence represents Q={Q _object, Q _attribute, Q _rest;

Step 3.3: for the situation comprising two and two or more domain object, first according to field concept object model, judge the relation between object, if between an object and other objects be the relation of object and attribute, then think that question sentence only comprises an object, all the other are the attribute of object, go to step 3.2; If the concept belonging to an object and the concept belonging to another object are hyponymies, then with the domain object that the next object is unique, go to step 3.2; Other situations go to step 3.4;

Step 3.4: situation question sentence being comprised to multiple domain object, is expressed as domain object, object properties, object relationship and other four parts question sentence, Q={Q _object, Q _attribute, Q _relation, Q _rest, obtain the conceptual object pattern representation of user's question sentence.

Preferably, Question sentence parsing in described step 4 calculates the Similarity Measure, remainder Similarity Measure four part that are decomposed between Similarity Measure between Similarity Measure between object that question sentence comprises, object properties, object relationship; Wherein the Similarity Measure of two corresponding set is all regarded in the calculating of each several part as, successively from one set choose an element gather with another respectively in element calculate similarity, pick out the element pair of maximum similarity, circulation is until first set is for empty; Then similarity right for these elements picked out is added, divided by the element number that first set comprises; Finally the result calculated based on two set respectively is on average obtained the similarity of two set, the Similarity Measure between question sentence is obtained by four some numerical results weighted calculation.

Preferably, the concrete steps of described step 4 are as follows:

Step 4.1: object part Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{obj}}(Q_{\mathrm{obj}}^{\&prime;},Q_{\mathrm{obj}})=\frac{1}{2}(\frac{1}{n_1}\underset{u=1}{\overset{n_1}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(O_u,O_v)\right)+\frac{1}{m_1}\underset{v=1}{\overset{m_1}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(O_u,O_v)\right))$

Sim (O in formula _u, O _v) be object O _uand O _vsimilarity, if be same object, its value is 1, otherwise is 0, O _uand O _vq' respectively _obj, Q _objin relation, n1 and m1 is Q' respectively _obj, Q _objthe number of middle relation;

Step 4.2: attribute section Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{att}}(Q_{\mathrm{att}}^{\&prime;},Q_{\mathrm{att}})=\frac{1}{2}(\frac{1}{n_2}\underset{u=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(A_u,A_v)\right)+\frac{1}{m_2}\underset{v=1}{\overset{m_2}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(A_u,A_v)\right))$

Sim (A in formula _u, A _v) be attribute A _uwith attribute A _vsimilarity, if be same attribute, its value is 1, otherwise is 0, attribute A _uwith attribute A _vq' respectively _att, Q _attin relation, n2 and m2 is Q' respectively _att, Q _attthe number of middle relation;

Step 4.3: the Similarity Measure between relation:

${\mathrm{SIMq}}_{\mathrm{rel}}(Q_{\mathrm{rel}}^{\&prime;},Q_{\mathrm{rel}})=\frac{1}{2}(\frac{1}{n_3}\underset{u=1}{\overset{n_3}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(R_u,R_v)\right)+\frac{1}{m_3}\underset{v=1}{\overset{m_3}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(R_u,R_v)\right))$

Sim (R in formula _u, R _v) be relation R _uwith relation R _vsimilarity, if be same relation, its value is 1 otherwise is 0, R _uand R _vq' respectively _rel, Q _relin relation, n3 and m3 is Q' respectively _rel, Q _relthe number of middle relation;

Step 4.4: remainder Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{res}}(Q_{\mathrm{res}}^{\&prime;},Q_{\mathrm{res}})=\frac{1}{2}(\frac{1}{n_4}\underset{u=1}{\overset{n_4}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(W_u,W_v)\right)+\frac{1}{m_4}\underset{v=1}{\overset{m_4}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(W_u,W_v)\right))$

Sim (W in formula _u, W _v) be word W _uwith word W _vsimilarity, its computing method based on semantic computation mode, W _uand W _vq' respectively _resand Q _resin word, n4 and m4 is Q' respectively _restand Q _restthe number of middle word;

Step 4.5: the Similarity Measure of final question sentence:

SIM(Q',Q)＝a ₁SIMq _obj(Q' _obj,Q _obj)+a ₂SIMq _att(Q' _att,Q _att)

+a ₃SIMq _rel(Q' _rel,Q _rel)+a ₄SIMq _res(Q' _res,Q _res)

Wherein, a ₁+ a ₂+ a ₃+ a ₄=1, a ₁, a ₂, a ₃, a ₄represent the partition factor of each similarity respectively.

Preferably, the theme of the object encoding question sentence in question sentence, the focus of its attribute representation's question sentence, wherein a ₁=0.3, a ₂=0.4, a ₃=0.15, a ₄=0.15.

Preferably, if lack certain part in the object model representation of question sentence, then the coefficient of this part is prorated in other coefficients.

Described question sentence retrieval, for retrieving the question sentence similar to user's question sentence, is also sorted by Similarity Measure and is obtained result for retrieval.

Described question sentence retrieval, for retrieving the question sentence similar to user's question sentence, is also sorted by Similarity Measure and is obtained result for retrieval.

Accompanying drawing explanation

Fig. 1 conceptual object model schematic

Fig. 2 Similarity Measure logical diagram

Embodiment

Specific embodiment of the invention is described in detail below for colleges and universities' enrolment consultation field; what be necessary to herein means out is; below implement just to further illustrate for of the present invention; limiting the scope of the invention can not be interpreted as; some nonessential improvement and adjustment that this art skilled person makes the present invention according to the invention described above content, still belong to protection scope of the present invention.

First defined declaration is carried out to the object in domain knowledge, concept, attribute etc.

Definition 1 to as if people to carry out put question to understand anything, it can be concrete things, also can be abstract event.

Defining 2 concepts, to have the abstract of the object of same characteristic features be exactly concept.

Define 3 attributes for describing a certain method characteristic of concept or object.

In colleges and universities' enrolment consultation field, " specialty " is exactly a concept, and " statistics ", " sociology " are exactly two objects, and " enrollment " is exactly the attribute of object " statistics ", " sociology ".

In colleges and universities' enrolment consultation field, artificial extraction concept wherein, object, attribute difference formalized description are as follows:

Concept set: Concepts={C ₁, C ₂... C _n(1)

Object set: Objects={O ₁, O ₂..., O _m(2)

Community set: Attributes={A ₁, A ₂..., A _k(3)

To use for reference between WordNet notional word and in HowNet justice former between relation (such as: hyponymy, synonymy, part-whole relationship etc.) design definition field concept between relation, and by the relation manually marked between concept.Obtain all set of relationship, formalization representation is as follows:

Set of relationship: Relations={R ₁, R ₂..., R _l(4)

In these relations, find the relation between concept and attribute, the inheritance between concept and concept by analysis, be related to that this three classes Relationship Comparison is special between concept and object, individually represent, remaining relation set of relationship Relations represents.

Using obtain above concept set, object set, community set, set of relationship is as concept node, Object node, attribute node, the relation node of conceptual object model.Wherein, rectangle represents concept node, round rectangle represents Object node, ellipse representation attribute node, rhombus represent relation node.Internodal line adopts following rule: connect with arrow line between upper subordinate concept, is connected, is connected, connects with straight line between all the other nodes between concept with example object with the line of 2 end band points between arrow points subordinate concept, concept with its attribute with the line of an end band point.The attribute of object can obtain from the concept belonging to it, does not describe in a model.Specifically see Figure of description 1.

The conceptual object model form of object and Attribute Recognition and question sentence represents that step is as follows:

Inquiry question sentence for user needs first to carry out word segmentation processing, and the present invention utilizes the Words partition system ICTCLAS of the Chinese Academy of Sciences, then removes stop words, obtains user and inquires about question sentence keyword and represent Q={W1, W2 ..., Wn}.Then identify the object in question sentence and attribute thereof, concrete recognizer is as follows:

Input: user's question sentence keyword represents Q={W1, W2 ..., Wn}.

Export: the object model of user's question sentence represents.

Step 1: utilize Forward Maximum Method method, finds out all spectra object in user's question sentence according to domain object storehouse, if question sentence is only containing a domain object, goes to step 2, otherwise goes to step 3.

Step 2: for domain object, according to field concept object model, obtain the attribute of attribute as object of its concept, the object properties utilizing maximum forward matching algorithm to find out user's question sentence to comprise, if do not find object properties in question sentence, then provide a default property.Such as object is " accounting speciality ", then its default property is " brief introduction " attribute.The conceptual object finally forming inquiry question sentence represents Q={Q _object, Q _attribute, Q _rest.

Step 3: for the situation comprising two or more domain object, formality, according to conceptual object model, judges the relation between object, if between an object and other objects be the relation of object and attribute, then think that question sentence only comprises an object, all the other are the attribute of object, go to step 2.If the concept belonging to an object and the concept belonging to another object are hyponymies, then with the domain object that the next object is unique, go to step 2.Other situations go to step 4.

Step 4: situation question sentence being comprised to multiple domain object, is expressed as domain object, object properties, object relationship and other four parts question sentence, Q={Q _object, Q _attribute, Q _relation, Q _rest.

By above-mentioned algorithm, obtain the object model that user inquires about question sentence and represent.

User's Similarity Measure inquired about between question sentence and candidate's question sentence collection is decomposed into Similarity Measure, remainder Similarity Measure four part between Similarity Measure between Similarity Measure between object that question sentence comprises, object properties, object relationship.

The Similarity Measure of two corresponding set is all regarded in the calculating of each several part as, successively from one set choose an element gather with another respectively in element calculate similarity, the element pair of the maximum similarity selected, circulation is until first set is for empty.Then similarity right for these elements picked out is added, divided by the element number that first set comprises.Finally the result calculated based on two set respectively is on average obtained the similarity of two set.Each several part Similarity Measure is specific as follows:

Step 4.1: object part Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{obj}}(Q_{\mathrm{obj}}^{\&prime;},Q_{\mathrm{obj}})=\frac{1}{2}(\frac{1}{n_1}\underset{u=1}{\overset{n_1}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(O_u,O_v)\right)+\frac{1}{m_1}\underset{v=1}{\overset{m_1}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(O_u,O_v)\right))$

Sim (O in formula _u, O _v) be object O _uand O _vsimilarity, if be same object, its value is 1, otherwise is 0, O _uand O _vq' respectively _obj, Q _objin relation, n and m is Q' respectively _obj, Q _objthe number of middle relation;

Step 4.2: attribute section Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{att}}(Q_{\mathrm{att}}^{\&prime;},Q_{\mathrm{att}})=\frac{1}{2}(\frac{1}{n_2}\underset{u=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(A_u,A_v)\right)+\frac{1}{m_2}\underset{v=1}{\overset{m_2}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(A_u,A_v)\right))$

Sim (A in formula _u, A _v) be attribute A _uwith attribute A _vsimilarity, if be same attribute, its value is 1, otherwise is 0.Attribute A _uwith attribute A _vq' respectively _att, Q _attin relation, n and m is Q' respectively _att, Q _attthe number of middle relation;

Step 4.3: the Similarity Measure between relation:

${\mathrm{SIMq}}_{\mathrm{rel}}(Q_{\mathrm{rel}}^{\&prime;},Q_{\mathrm{rel}})=\frac{1}{2}(\frac{1}{n_3}\underset{u=1}{\overset{n_3}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(R_u,R_v)\right)+\frac{1}{m_3}\underset{v=1}{\overset{m_3}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(R_u,R_v)\right))$

Sim (R in formula _u, R _v) be relation R _uwith relation R _vsimilarity, if be same relation, its value is 1 otherwise is 0.R _uand R _vq' respectively _rel, Q _relin relation, n and m is Q' respectively _rel, Q _relthe number of middle relation;

Step 4.4: remainder Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{res}}(Q_{\mathrm{res}}^{\&prime;},Q_{\mathrm{res}})=\frac{1}{2}(\frac{1}{n_4}\underset{u=1}{\overset{n_4}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(W_u,W_v)\right)+\frac{1}{m_4}\underset{v=1}{\overset{m_4}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(W_u,W_v)\right))$

Sim (W in formula _u, W _v) be word W _uwith word W _vsimilarity, its computing method are based on knowing net semantic computation, W _uand W _vq' respectively _resand Q _resin word, n and m is Q' respectively _restand Q _restthe number of middle word;

Step 4.5: the similarity calculating method between final question sentence is as follows

SIM(Q',Q)＝a ₁SIMq _obj(Q' _obj,Q _obj)+a ₂SIMq _att(Q' _att,Q _att)

+a ₃SIMq _rel(Q' _rel,Q _rel)+a ₄SIMq _res(Q' _res,Q _res)

Wherein, a ₁+ a ₂+ a ₃+ a ₄=1, find through research, the theme of the object encoding question sentence in question sentence, the focus of its attribute representation's question sentence, these two parts describe the main aspect that user puts question to intention, and user more pays close attention to certain aspect attribute of object, therefore arranges a herein ₁=0.3, a ₂=0.4, a ₃=0.15, a ₄=0.15, if lack certain part in the object model representation of question sentence, then the coefficient of this part is prorated in other coefficients.

The algorithm steps of question sentence retrieval.

In order to improve question sentence effectiveness of retrieval, question and answer are expressed as conceptual object model all in advance to the question sentence concentrated, and concrete question sentence searching algorithm is as follows:

Input: user inquires about question sentence

Export: the most similar front 5 question sentences

Step 1: user inquires about question sentence participle, stop words process and synonym process;

Step 2: user inquires about question sentence object and attribute extraction thereof;

Step 3: calculate the similarity that user inquires about question sentence and candidate's question sentence, if similarity is greater than the threshold value of setting, then this question sentence is put into candidate result and concentrate, then calculates the similarity with next candidate's question sentence, until candidate's question sentence collection is empty.

Step 4: the candidate result collection obtained is arranged from big to small by similarity, gets front 5 question sentences as the result returned.

Although for illustrative purposes; describe illustrative embodiments of the present invention; but it should be appreciated by those skilled in the art that; when not departing from scope of invention disclosed in claims and spirit; the change of various amendment, interpolation and replacement etc. can be carried out in form and details; and all these change the protection domain that all should belong to claims of the present invention; and application claims protection each department of product and method in each step, can combine with the form of combination in any.Therefore, be not intended to limit the scope of the invention to the description of embodiment disclosed in the present invention, but for describing the present invention.Correspondingly, scope of the present invention not by the restriction of above embodiment, but is limited by claim or its equivalent.

Claims (8)

1. a question sentence search method for concept based object model, is characterized in that, in turn include the following steps:

Step 1: according to the feature of research restriction field question sentence, according to conceptual object model knowledge representation method, set up field concept object model;

Step 2: user input text question sentence is also determined;

Step 3: character crossfire process, extracts concept, object and the attribute in question sentence, utilizes the theme of concept or object factory question sentence, utilize the attribute description question sentence focus of concept or object, question sentence is expressed as conceptual object model representation form;

Step 4: the Similarity Measure mode of concept based object model definition question sentence, be specially and Question sentence parsing calculating is divided into four parts, Similarity Measure and remainder Similarity Measure between Similarity Measure, relation between Similarity Measure, attribute between object, last weighted calculation obtains the Similarity Measure mode of question sentence;

Step 5: the Similarity Measure mode based on question sentence carries out question sentence retrieval.

2. the question sentence search method of a kind of concept based object model as claimed in claim 1, it is characterized in that: described according to conceptual object model knowledge representation method, set up field concept object model, concrete steps are: therefrom extract field concept by analysis field question sentence feature, domain object, concept attribute and object properties, and the relation between them, obtain field concept set, domain object set, community set and set of relationship, the representation of the contextual definition relation by analysis in set of relationship, relation wherein between concept and attribute, inheritance between concept and concept, relation between concept and object individually represents, finally set up field concept object model, specific as follows:

Step 1.1: extract field concept, set up field concept set Concepts={C ₁, C ₂... C _n;

Step 1.2: extract domain object, set up domain object set Objects={O ₁, O ₂..., O _m;

Step 1.3: extract concept attribute and object properties, set up community set Attributes={A ₁, A ₂..., A _k;

Step 1.4: extract field concept, domain object, relation between concept attribute and object properties, opening relationships set Relations={R ₁, R ₂..., R _l;

Step 1.5: the representation of defined notion, the relation between attribute and object, the relation between concept and attribute, the inheritance between concept and concept, the relation between concept and object and other relations, sets up field concept object model.

3. the question sentence search method of a kind of concept based object model as claimed in claim 1, is characterized in that: the concrete steps of described step 3 are as follows:

Step 3.1: utilize Forward Maximum Method method, finds out all spectra object in user's question sentence according to domain object storehouse, if question sentence is only containing a domain object, goes to step 3.2, otherwise goes to step 3.3;

Step 3.2: for domain object, according to field concept object model, obtain the attribute of its concept as object properties, the object properties utilizing Forward Maximum Method method to find out user's question sentence to comprise, if do not find object properties in question sentence, then provide a default property, the conceptual object finally forming inquiry question sentence represents Q={Q _object, Q _attribute, Q _rest;

Step 3.3: for the situation comprising two and two or more domain object, first according to field concept object model, judge the relation between object, if between an object and other objects be the relation of object and attribute, then think that question sentence only comprises an object, all the other are the attribute of object, go to step 3.2; If the concept belonging to an object and the concept belonging to another object are hyponymies, then with the domain object that the next object is unique, go to step 3.2; Other situations go to step 3.4;

Step 3.4: situation question sentence being comprised to multiple domain object, is expressed as domain object, object properties, object relationship and other four parts question sentence, Q={Q _object, Q _attribute, Q _relation, Q _rest, obtain the conceptual object pattern representation of user's question sentence.

4. the question sentence search method of a kind of concept based object model as claimed in claim 1, is characterized in that: Question sentence parsing in described step 4 calculates the Similarity Measure, remainder Similarity Measure four part that are decomposed between Similarity Measure between Similarity Measure between object that question sentence comprises, object properties, object relationship; Wherein the Similarity Measure of two corresponding set is all regarded in the calculating of each several part as, successively from one set choose an element gather with another respectively in element calculate similarity, pick out the element pair of maximum similarity, circulation is until first set is for empty; Then similarity right for these elements picked out is added, divided by the element number that first set comprises; Finally the result calculated based on two set respectively is on average obtained the similarity of two set, the Similarity Measure between question sentence is obtained by four some numerical results weighted calculation.

5. the question sentence search method of a kind of concept based object model as described in claim 1 or 4, is characterized in that: the concrete steps of described step 4 are as follows:

Step 4.1: object part Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{obj}}(Q_{\mathrm{obj}}^{\&prime;},Q_{\mathrm{obj}})=\frac{1}{2}(\frac{1}{n_1}\underset{u=1}{\overset{n_1}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(O_u,O_v)\right)+\frac{1}{m_1}\underset{v=1}{\overset{m_1}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(O_u,O_v)\right))$

Sim (O in formula _u, O _v) be object O _uand O _vsimilarity, if be same object, its value is 1, otherwise is 0, O _uand O _vq' respectively _obj, Q _objin relation, n1 and m1 is Q' respectively _obj, Q _objthe number of middle relation;

Step 4.2: attribute section Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{att}}(Q_{\mathrm{att}}^{\&prime;},Q_{\mathrm{att}})=\frac{1}{2}(\frac{1}{n_2}\underset{u=1}{\overset{n_2}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(A_u,A_v)\right)+\frac{1}{m_2}\underset{v=1}{\overset{m_2}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(A_u,A_v)\right))$

Sim (A in formula _u, A _v) be attribute A _uwith attribute A _vsimilarity, if be same attribute, its value is 1, otherwise is 0, attribute A _uwith attribute A _vq' respectively _att, Q _attin relation, n2 and m2 is Q' respectively _att, Q _attthe number of middle relation;

Step 4.3: the Similarity Measure between relation:

${\mathrm{SIMq}}_{\mathrm{rel}}(Q_{\mathrm{rel}}^{\&prime;},Q_{\mathrm{rel}})=\frac{1}{2}(\frac{1}{n_3}\underset{u=1}{\overset{n_3}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(R_u,R_v)\right)+\frac{1}{m_3}\underset{v=1}{\overset{m_3}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(R_u,R_v)\right))$

Sim (R in formula _u, R _v) be relation R _uwith relation R _vsimilarity, if be same relation, its value is 1 otherwise is 0, R _uand R _vq' respectively _rel, Q _relin relation, n3 and m3 is Q' respectively _rel, Q _relthe number of middle relation;

Step 4.4: remainder Similarity Measure:

${\mathrm{SIMq}}_{\mathrm{res}}(Q_{\mathrm{res}}^{\&prime;},Q_{\mathrm{res}})=\frac{1}{2}(\frac{1}{n_4}\underset{u=1}{\overset{n_4}{\mathrm{\&Sigma;}}}\underset{1<v<m}{\max }\left(\mathrm{sim}(W_u,W_v)\right)+\frac{1}{m_4}\underset{v=1}{\overset{m_4}{\mathrm{\&Sigma;}}}\underset{1<u<n}{\max }\left(\mathrm{sim}(W_u,W_v)\right))$

Sim (W in formula _u, W _v) be word W _uwith word W _vsimilarity, its computing method based on semantic computation mode, W _uand W _vq' respectively _resand Q _resin word, n4 and m4 is Q' respectively _restand Q _restthe number of middle word;

Step 4.5: the Similarity Measure of final question sentence:

SIM(Q',Q)＝a ₁SIMq _obj(Q' _obj,Q _obj)+a ₂SIMq _att(Q' _att,Q _att)

+a ₃SIMq _rel(Q' _rel,Q _rel)+a ₄SIMq _res(Q' _res,Q _res)

Wherein, a ₁+ a ₂+ a ₃+ a ₄=1, a ₁, a ₂, a ₃, a ₄represent the partition factor of each similarity respectively.

6. the question sentence search method of a kind of concept based object model as claimed in claim 5, is characterized in that: the theme of the object encoding question sentence in question sentence, the focus of its attribute representation's question sentence, wherein a ₁=0.3, a ₂=0.4, a ₃=0.15, a ₄=0.15.

7. the question sentence search method of a kind of concept based object model as claimed in claim 5, is characterized in that: if lack certain part in the object model representation of question sentence, then the coefficient of this part is prorated in other coefficients.

8. the question sentence search method of a kind of concept based object model as claimed in claim 1, is characterized in that: described question sentence retrieval, for retrieving the question sentence similar to user's question sentence, is also sorted by Similarity Measure and obtained result for retrieval.