CN101119326B - Method and device for managing instant communication conversation record - Google Patents

Abstract

The present invention discloses a managing method for instant communication conversation records, which aims at solving the prior problems that inquiring information in the conversation records is not only fussy but also inefficiency for instant communicating users. The method comprises the following steps: getting and assorting the conversation records of users to get sample groups; correlating analyzing each sample group to get a corresponding assorted group that contains eigenvector corresponding with the conversation records in the sample groups; determining the conversation topic corresponding with the assorted group according to the emerging frequency of the words in the assorted groups, and relating the conversation topic to the conversation records corresponding with the assorted group; searching for the conversation topic that matches with the key word inputted by the users, and displaying the conversation records that relates with the conversation topic to the users. The present invention also discloses a managing device used for instant communication conversation records.

Description

A kind of management method of instant communication conversation recording and device

Technical field

The present invention relates to communication and field of computer technology, relate in particular to a kind of management method and device of instant telecommunication session record.

Background technology

Along with the continuous development of instant messaging (IM) technology with popularize, increasing user not only adopts IM software to exchange with other users in network, the instrument that IM software can also be encountered problems in other user's works of consultation or study as the user, simultaneously, conversation recording between the user is accompanied by under interchange between the user preserves in the IM system, provides data for the user searches the information of oneself paying close attention to later on.

For example: when user A seeks advice from a problem to user B, user B has returned the answer of problem, as user C during with regard to same problem counsel user A or user B, user A need check the relevant information in the conversation recording with user B, when perhaps user B need check relevant information in the conversation recording with user A, user A or user B need manually search relative recording in conversation recording, the time interval more when conversation recording or user A and user C counseling problem is when longer, adopt the method for prior art, not only increased the workload of manually searching, and search efficiency is lower.

If user A seeks advice from a plurality of users with regard to same problem, when user A wish from a plurality of users' conversation recording during Query Information, adopt the method for prior art, when the instantaneous communication system that uses as the user provides the instantaneous communication system of conversation recording look facility, user A can only manually check a plurality of users' conversation recording one by one, finds the information of oneself being concerned about.Even user A uses some other that instantaneous communication system of the data importing/export function of user conversation record is provided, user A also needs a plurality of users' conversation recording data are derived earlier, in derived data, inquire about then, user A also can inquire about in derived data according to the keyword of the information of oneself being concerned about, but adopt the mode of keyword also can only navigate to the paragraph that comprises this keyword, this paragraph is not necessarily relevant with the information that the user is concerned about, can not realize that the user effectively searches information in conversation recording.

Summary of the invention

The invention provides a kind of management method and device of instant telecommunication session record, in order to solve the instant communication user that exists in the prior art in conversation recording during Query Information, the problem that not only complex operation, and search efficiency is low.

The invention provides following technical scheme:

A kind of management method of instant communication conversation recording comprises the steps:

Obtain user's conversation recording and it is classified and obtain sample set;

Generate every conversation recording characteristic of correspondence vector in the described sample set, analyze the correlation of each characteristic vector and other characteristic vectors, according to described correlation to the characteristic vector generation sort merge of classifying;

Determine the session theme of sort merge correspondence according to the frequency that word in each sort merge occurs, and make this session theme be associated with the conversation recording of sort merge correspondence; And

The session theme of the keyword lookup of importing during according to user inquiring and this keyword coupling, and the conversation recording related with the session theme that will find presented to the user.

Wherein, correlation behind the generation session theme between the further analysis session theme, and correlation merged into same session theme greater than the session theme of predetermined threshold, make the session theme after the merging related with the pairing conversation recording of merged all session themes.

By different session subscriber to the conversation recording generation sample set of classifying.

Preferable, according to the blanking time of conversation recording in the described sample set, further a sample set is divided into a plurality of different sample sets.

Generate every conversation recording characteristic of correspondence vector in the sample set, analyze the correlation of each characteristic vector and other characteristic vectors, specifically comprise:

Every conversation recording is carried out word segmentation processing, and the word of deleting no practical significance in this conversation recording obtains S set, merges the synonym among the S, and carries out vectorization, then generates and this conversation recording characteristic of correspondence vector

(W ₁, W ₂, W ₃... Wn), wherein Wi is the weight of i element, and each element is the word among the S;

Calculate and each conversation recording characteristic of correspondence vector

In the weight of each speech, according to the correlation of weight calculation each characteristic vector of each speech in its characteristic vector of forming described characteristic vector.

Determine the session theme of this sort merge greater than the word of predetermined threshold according to the frequency of occurrences in the sort merge.

A kind of management devices of instant communication conversation recording comprises:

Be used to store the unit of user conversation record;

Be used for described conversation recording classified and generate the unit of sample set;

Be used for generating every conversation recording characteristic of correspondence of described sample set vector, analyze the correlation of each characteristic vector and other characteristic vectors, according to described correlation characteristic vector being classified generates the unit of sort merge;

Be used for determining the session theme of described sort merge correspondence, and make this session theme be associated with the unit of the conversation recording of sort merge correspondence; And

The session theme of the keyword lookup of importing when being used for and this keyword coupling, and the conversation recording related that will the find unit of presenting to the user with the session theme according to user inquiring.

Preferable, described device also comprises:

Be used for the correlation between the analysis session theme, and correlation is merged into same session theme greater than the session theme of predetermined threshold, and session theme after will merging and the related unit of the merged pairing conversation recording of all session themes.

Beneficial effect of the present invention is as follows:

The present invention classifies to the user conversation record behind the generation sample set, respectively each sample set is carried out the session theme that correlation analysis generates the respective classified combination and determines the sort merge correspondence, and the conversation recording that the session theme is associated with the sort merge correspondence.After adopting the present invention, when the user need be from conversation recording during Query Information, the user only need import keyword, system will search the session theme with this keyword coupling automatically, and the associated conversation recording of session theme that finds presented to the user, troublesome operation when not only having avoided user's craft Query Information, and improved search efficiency.

Description of drawings

Fig. 1 is the management devices structural representation of user conversation record in the embodiment of the invention;

Fig. 2 is the schematic diagram of user conversation record management method in the embodiment of the invention;

The process chart of Fig. 3 in the embodiment of the invention user conversation record being classified;

Fig. 4 is for carrying out the process chart of correlation analysis to sample set in the embodiment of the invention.

Embodiment

In order to solve in the prior art, instant communication user is in conversation recording during Query Information, complex operation not only, and the low problem of search efficiency, in the present embodiment user conversation is write down the generation sample set of classifying, respectively each sample set is carried out the session theme that correlation analysis generates the respective classified combination and determines the sort merge correspondence, and the session theme is associated with the conversation recording of sort merge correspondence, and the session theme that mates according to the keyword lookup of user input and this keyword, and the associated conversation recording of session theme that finds presented to the user.

Consult the management devices structural representation that Figure 1 shows that user conversation record in the present embodiment, comprising: memory cell 101, taxon 102, analytic unit 103, session thematic unit 104, merge cells 105 and query unit 106.

Memory cell 101 is used to preserve user's conversation recording and session theme.Taxon 102 is used to obtain conversation recording and conversation recording is classified and obtains sample set.Analytic unit 103 is used for sample set is carried out correlation analysis, generates the sort merge of sample set.Session thematic unit 104 is used for determining the session theme of sample set sort merge, and makes this session theme be associated with the conversation recording of sort merge correspondence.Merge cells 105 is used for the correlation between the analysis session theme, and correlation is merged into same session theme greater than the session theme of predetermined threshold, and the session theme after will merging is associated with the conversation recording of merged all session theme correspondences.Query unit 106 is used for receiving the keyword that the user imports and searches session theme with this keyword coupling when the conversation recording Query Information, and the associated conversation recording of session theme that finds is presented to the user.

Consult the schematic diagram that Figure 2 shows that user conversation record management method in the present embodiment, comprising:

Step 201, obtain user's conversation recording and this conversation recording classified and obtain sample set.

Step 202, the sample set that generates is carried out correlation analysis generate the respective classified combination.

Step 203, the frequency that occurs according to word in each sort merge are determined the session theme of sort merge correspondence, and are made this session theme be associated with the conversation recording of sort merge correspondence.

Correlation between step 204, the analysis session theme, and correlation merged into same session theme greater than the session theme of predetermined threshold makes session theme after the merging be associated with the conversation recording of merged all session theme correspondences.

Step 205, when the user in conversation recording during Query Information, the session theme of the keyword lookup of importing during according to user inquiring and this keyword coupling, and the associated conversation recording of session theme that finds presented to the user.

In step 201, the handling process that conversation recording is classified is consulted shown in Figure 3, and processing procedure is as follows:

Step 301, judge that whether conversation recording handle through classification,, then it is not handled if handle through classification; Otherwise, execution in step 302.

Step 302, the conversation recording of handling through classification is not classified to conversation recording according to different users, as: judge conversation recording TRi and conversation recording TR _jWhether belong to the conversation recording between same user, if conversation recording TRi and conversation recording TR _jBelong to the session between different user, TRi and conversation recording TR are write down in session _jBe divided into different sample set TS; If conversation recording TRi and conversation recording TR _jBelong to the conversation recording between same user, then session is write down TRi and conversation recording TR _jBe divided in the identical sample set.

Step 303, with same sample set according to dividing the blanking time of the conversation recording in this sample set, further be divided into different sample sets, can be made as a week etc. according to practical application the blanking time of conversation recording.

Handling the sample set TS that generates through step 303 is the sample set that carries out correlation analysis.

Consult shown in Figure 4ly, it is as follows to adopt KNN (K Nearest Neighbor, K nearest-neighbors) algorithm to carry out the processing procedure of correlation analysis to sample set:

Step 401, every conversation recording TR among the sample set TS is generated the characteristic of correspondence vector.At first every conversation recording TR is carried out word segmentation processing, remove auxiliary word wherein, the speech of no practical significance such as interjection obtains S set; Merge the synonym among the S, for example { " computer ", " computer " } merged into { " computer ", " computer " }.The S set corresponding to every conversation recording after will merging through synonym is carried out vectorization, the generating feature vector

(W ₁, W ₂, W ₃... Wn), wherein Wi is the weight of i element, and each element is the word among the S.

Step 402, calculating and each conversation recording TR characteristic of correspondence vector

In the weights W of each element.Adopting following formula to carry out weights calculates:

$W(t,\stackrel{\→}{d})=\frac{\mathrm{tf}(t,\stackrel{\→}{d})\×\log (N/n_t+0.01)}{\sqrt{{\mathrm{\Σ}}_{t\∈\stackrel{\→}{d}}[\mathrm{tf}(t,\stackrel{\→}{d})\×\log (N{/n}_t+0.01){]}^2}}$

Wherein,

For speech t in characteristic vector

In weight, and For speech t in characteristic vector In word frequency, N is the sum of conversation recording TR among each sample set TS, n _tFor occurring the conversation recording TR number of speech t among each sample set TS, denominator is a normalization factor.

Coefficient correlation between step 403, calculating and each conversation recording characteristic of correspondence vector is determined K the characteristic vector the most similar to each characteristic vector according to calculating the gained coefficient correlation.

During concrete enforcement, adopt following formula $\mathrm{Sim}(d_i,d_j)=\frac{\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{W}_{\mathrm{ik}}\×W_{\mathrm{jk}}}{\sqrt{\left(\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{W}_{\mathrm{ik}}^2\right)\left(\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{W}_{\mathrm{jk}}^2\right)}}$ Calculate the coefficient correlation between each conversation recording characteristic of correspondence vector, wherein, Sim (d _i, d _j) be characteristic vector d _iWith characteristic vector d _jCoefficient correlation, W _IkAnd W _JkBe respectively characteristic vector d _iWith characteristic vector d _jThe weights of k element.

By calculating, obtain the coefficient correlation between each characteristic vector, according to this coefficient correlation, will be combined as a set respectively with the maximally related K of each a characteristic vector characteristic vector, the value of K can be determined according to practical application.

Step 404, each conversation recording characteristic of correspondence vector is divided in the inhomogeneity of classification among the C generates sort merge.

Classification C is the set that each conversation recording characteristic of correspondence vector is formed among the sample set TS.

Method one: when classification C is sky, then generate a vector set c among the classification C in the following way, then c is added among the C that classifies:

Characteristic vector corresponding to conversation recording

And characteristic vector

Belong to the set that the most similar K of the other side neighbours form respectively, then

With

Belong to same class c, generate class c and such and characteristic vector

And characteristic vector Classification C is added class c in corresponding conversation recording association then, and the characteristic vector among each class c is formed a sort merge.

Method two: when classification C is not sky, then calculate characteristic vector corresponding to each conversation recording

Belong to the weight of certain class c (c ∈ C), adopt following formula:

$p(\stackrel{\→}{x},C_j)=\underset{{\stackrel{\→}{d}}_i\∈\mathrm{KNN}}{\mathrm{\Σ}}\mathrm{Sim}(\stackrel{\→}{x},{\stackrel{\→}{d}}_i)y({\stackrel{\→}{d}}_i,C_j)$

Wherein,

Be characteristic vector corresponding to a conversation recording,

For with

Characteristic vector in the set of the most similar K neighbours' composition,

For

Characteristic vector maximally related with it

Coefficient correlation, this coefficient correlation can obtain according to step 403 result of calculation,

Be the category attribute function, if characteristic vector

Belong to class C _j,

Functional value be 1, otherwise be 0.According to calculating The comparative feature vector

At all kinds of C _jIn weights, with characteristic vector

Assign to the bigger class C of weights _jIn, and with such C _jWith characteristic vector

Corresponding conversation recording association.

When adopting method two, if characteristic vector

All very little with the degree of correlation of existing each class c, then can adopt the mode of method one to generate a new class c ', and class c ' is joined among the classification C, and with class c ' and eigen vector

Corresponding conversation recording association.

After each characteristic vector handled, characteristic vector all is divided in the class, by all kinds of sort merges of forming respectively.

N word that the frequency of occurrences in each sort merge that generates is the highest or frequency are defined as the session theme of this sort merge greater than the word of a, and N value and a value are determined according to practical application.

Each sample set TS is carried out generating after the above-mentioned processing session theme of sort merge and this sort merge correspondence, when the session theme that generates is carried out correlation analysis, with the sample set of session theme as the KNN algorithm, calculate in this set the weight of each speech in this session theme in each session theme, according to weight, utilize the formula in the step 403, calculate the coefficient correlation of each session theme, the session theme of coefficient correlation greater than setting threshold merged.

When presenting conversation recording, the session record is arranged, also can be arranged according to the weight order of conversation recording in the session theme according to different session subscriber to the user.

Adopted the KNN algorithm that sample set is carried out correlation analysis among the above embodiment, but the present invention is not limited only to adopt the KNN algorithm that sample set is analyzed.Conversation recording is carried out the method for correlation analysis and can also use training algorithm and the sorting technique based on vector space such as vector machine algorithm, neural network algorithm and bayesian algorithm.When for example adopting bayesian algorithm, calculate each speech in each conversation recording character pair vector and appear at probability in certain session, calculate the probability that characteristic vector belongs to certain session according to Bayesian formula then, it is joined in the session of probability maximum.

Adopt the present invention, when user's information that inquiring user is concerned about in conversation recording, the user only need import keyword, system will inquire about the session theme with the keyword coupling automatically, and the conversation recording related with this session theme presented to the user, troublesome operation when not only having avoided user's craft Query Information, and improved search efficiency.

Obviously, those skilled in the art can carry out various changes and modification to the present invention and not break away from the spirit and scope of the present invention.Like this, if of the present invention these are revised and modification belongs within the scope of claim of the present invention and equivalent technologies thereof, then the present invention also is intended to comprise these changes and modification interior.

Claims (8)

1. the management method of an instant communication conversation recording is characterized in that, comprises the steps:

Obtain user's conversation recording and it is classified and obtain sample set;

Generate every conversation recording characteristic of correspondence vector in the described sample set, analyze the correlation of each characteristic vector and other characteristic vectors, according to described correlation to the characteristic vector generation sort merge of classifying;

Determine the session theme of sort merge correspondence according to the frequency that word in each sort merge occurs, and make this session theme be associated with the conversation recording of sort merge correspondence; And

The session theme of the keyword lookup of importing during according to user inquiring and this keyword coupling, and the conversation recording related with the session theme that will find presented to the user.

2. the method for claim 1, it is characterized in that, correlation behind the generation session theme between the further analysis session theme, and correlation merged into same session theme greater than the session theme of predetermined threshold, make the session theme after the merging related with the pairing conversation recording of merged all session themes.

3. method as claimed in claim 1 or 2 is characterized in that, by different session subscriber to the conversation recording generation sample set of classifying.

4. method as claimed in claim 3 is characterized in that, according to the blanking time of conversation recording in the described sample set, further a sample set is divided into a plurality of different sample sets.

5. the method for claim 1 is characterized in that, every conversation recording characteristic of correspondence vector in the described generation sample set is analyzed the correlation of each characteristic vector and other characteristic vectors, specifically comprises:

Every conversation recording is carried out word segmentation processing, and the word of deleting no practical significance in this conversation recording obtains S set, merges the synonym among the S, and carries out vectorization, then generates and this conversation recording characteristic of correspondence vector

(W ₁, W ₂, W ₃... Wn), wherein Wi is the weight of i element, and each element is the word among the S;

Calculate and each conversation recording characteristic of correspondence vector

In the weight of each speech, according to the correlation of weight calculation each characteristic vector of each speech in its characteristic vector of forming described characteristic vector.

6. the method for claim 1 is characterized in that, determines the session theme of this sort merge greater than the word of predetermined threshold according to the frequency of occurrences in the sort merge.

7. the management devices of an instant communication conversation recording is characterized in that, comprising:

Be used to store the unit of user conversation record;

Be used for described conversation recording classified and generate the unit of sample set;

Be used for generating every conversation recording characteristic of correspondence of described sample set vector, analyze the correlation of each characteristic vector and other characteristic vectors, according to described correlation characteristic vector being classified generates the unit of sort merge;

Be used for determining the session theme of described sort merge correspondence, and make this session theme be associated with the unit of the conversation recording of sort merge correspondence; And

The session theme of the keyword lookup of importing when being used for and this keyword coupling, and the conversation recording related that will the find unit of presenting to the user with the session theme according to user inquiring.

8. device as claimed in claim 7 is characterized in that, also comprises:

Be used for the correlation between the analysis session theme, and correlation is merged into same session theme greater than the session theme of predetermined threshold, and session theme after will merging and the related unit of the merged pairing conversation recording of all session themes.

Images