CN101119326B - A management method and device for instant messaging session records - Google Patents

Abstract

The invention discloses a management method of an instant communication session record, which is used to solve the problem in the prior art that when an instant communication user queries information in the session record, not only the operation is cumbersome, but also the query efficiency is low. The method includes: obtaining user session records and classifying them to obtain a sample set; performing correlation analysis on each sample set to generate a corresponding classification combination, the classification combination including feature vectors corresponding to the session records in the sample set; The frequency of occurrence of words in each classification combination determines the corresponding conversation theme of the classification combination, and makes the conversation theme associated with the corresponding conversation record of the classification combination; Present the found session record associated with the session topic to the user. The invention also discloses a management device for instant communication session records.

Description

A management method and device for instant messaging session records

technical field

The invention relates to the technical fields of communication and computer, in particular to a method and device for managing instant messaging session records.

Background technique

With the continuous development and popularization of instant messaging (IM) technology, more and more users not only use IM software to communicate with other users in the network, but also use IM software as a user to consult other users encountered in work or study. At the same time, the conversation records between users are saved in the IM system along with the communication between users, which provides information for users to find the information they care about in the future.

For example: when user A consults user B about a question, and user B returns the answer to the question, when user C consults user A or user B on the same question, user A needs to check the session record with user B related information, or when user B needs to check the relevant information in the conversation records with user A, user A or user B needs to manually search for relevant records in the conversation records. When the time interval is long, adopting the method of the prior art not only increases the workload of manual search, but also has low search efficiency.

If user A has consulted multiple users on the same issue, when user A wishes to inquire information from the session records with multiple users, the method of the prior art is adopted, such as the instant messaging system used by the user to provide the session record viewing function In the instant messaging system, user A can only manually check the session records of multiple users one by one to find the information he cares about. Even if user A uses some other instant messaging system that provides the data import/export function of user session records, user A also needs to export the session record data of multiple users first, and then query in the exported data. User A can also According to the keywords of the information you care about, you can query in the exported data, but the method of using keywords can only locate the segment containing the keyword, and the segment is not necessarily related to the information that the user cares about, nor can the user Efficiently look up information in session recordings.

Contents of the invention

The present invention provides a management method and device for instant messaging session records, which are used to solve the problem of cumbersome operations and low query efficiency when instant messaging users query information in session records in the prior art.

The invention provides the following technical solutions:

A method for managing instant messaging session records, comprising the steps of:

Obtain the user's session records and classify them to obtain a sample set;

Generate a feature vector corresponding to each session record in the sample set, analyze the correlation between each feature vector and other feature vectors, and classify the feature vectors according to the correlation to generate a classification combination;

Determine the conversation topic corresponding to the classification combination according to the frequency of occurrence of words in each classification combination, and make the conversation topic associated with the corresponding conversation record of the classification combination; and

According to the keyword entered by the user when querying, search for a conversation topic matching the keyword, and present the found conversation record associated with the conversation topic to the user.

Wherein, after generating the session topics, further analyze the correlation between the session topics, and merge the session topics whose correlations are greater than a predetermined threshold into the same session topic, so that the merged session topics and the sessions corresponding to all the session topics that are merged Record association.

Classify session records by different session users to generate sample sets.

Preferably, a sample set is further divided into multiple different sample sets according to the interval time of session records in the sample set.

Generate feature vectors corresponding to each session record in the sample set, and analyze the correlation between each feature vector and other feature vectors, including:

(W₁，W₂，W₃......Wn)，其中Wi为第i个元素的权重，各元素为S中的词语；Perform word segmentation for each session record, delete meaningless words in the session record to obtain a set S, merge synonyms in S, and perform vectorization, then generate a feature vector corresponding to the session record

(W ₁ , W ₂ , W ₃ ......Wn), where Wi is the weight of the i-th element, and each element is a word in S;

中各词的权重，根据组成所述特征向量的各词在其特征向量中的权重计算各特征向量的相关性。Compute the feature vector corresponding to each session record

According to the weight of each word in the feature vector, the correlation of each feature vector is calculated according to the weight of each word forming the feature vector in its feature vector.

The conversation topic of the category combination is determined according to the words whose frequency of occurrence in the category combination is greater than a predetermined threshold.

A management device for instant messaging session records, comprising:

A unit for storing user session records;

A unit for classifying the session records to generate a sample set;

A unit for generating a feature vector corresponding to each session record in the sample set, analyzing the correlation between each feature vector and other feature vectors, and classifying the feature vectors according to the correlation to generate a classification combination;

A unit for determining a conversation topic corresponding to the classification combination, and associating the conversation topic with a conversation record corresponding to the classification combination; and

A unit for finding a conversation topic matching the keyword according to a keyword entered by the user when querying, and presenting the found conversation record associated with the conversation topic to the user.

Preferably, the device also includes:

A unit for analyzing the correlation between session topics, merging session topics whose correlation is greater than a predetermined threshold into the same session topic, and associating the merged session topic with the session records corresponding to all the merged session topics .

The beneficial effects of the present invention are as follows:

The present invention classifies user session records to generate sample sets, performs correlation analysis on each sample set to generate corresponding classification combinations, determines conversation topics corresponding to the classification combinations, and associates the conversation topics with session records corresponding to the classification combinations. After adopting the present invention, when the user needs to query information from the session record, the user only needs to input a keyword, and the system will automatically search for a session topic matching the keyword, and present the session record associated with the found session topic to the It not only avoids the cumbersome operation when users manually query information, but also improves the query efficiency.

Description of drawings

FIG. 1 is a schematic structural diagram of a management device for user session records in an embodiment of the present invention;

2 is a schematic diagram of a user session record management method in an embodiment of the present invention;

Fig. 3 is the processing flowchart of classifying user session records in the embodiment of the present invention;

Fig. 4 is a processing flow chart of performing correlation analysis on sample sets in an embodiment of the present invention.

Detailed ways

In order to solve the problem of cumbersome operations and low query efficiency in the prior art when instant messaging users query information in session records, in this embodiment, user session records are classified to generate sample sets, and each sample set is correlated Generative analysis generates the corresponding classification combination and determines the conversation topic corresponding to the classification combination, and associates the conversation topic with the conversation record corresponding to the classification combination, and finds the conversation topic matching the keyword according to the keyword input by the user, and searches The session record associated with the received session topic is presented to the user.

Referring to FIG. 1 , it is a schematic structural diagram of a management device for user session records in this embodiment, including: a storage unit 101 , a classification unit 102 , an analysis unit 103 , a session topic unit 104 , a merge unit 105 and a query unit 106 .

The storage unit 101 is used for saving user's session records and session topics. The classifying unit 102 is configured to acquire session records and classify the session records to obtain a sample set. The analysis unit 103 is used for performing correlation analysis on the sample set to generate a classification combination of the sample set. The conversation theme unit 104 is configured to determine the conversation theme of the classification combination of the sample set, and associate the conversation theme with the conversation record corresponding to the classification combination. The merging unit 105 is used for analyzing the correlation between the conversation topics, and merging the conversation topics whose correlation is greater than a predetermined threshold into the same conversation topic, and associating the merged conversation topic to the conversation records corresponding to all the conversation topics being merged . The query unit 106 is configured to receive a keyword input by the user when querying information in the conversation record, find a conversation topic matching the keyword, and present the conversation record associated with the found conversation topic to the user.

Referring to Figure 2, it is a schematic diagram of a user session record management method in this embodiment, including:

Step 201. Obtain the user's session records and classify the session records to obtain a sample set.

Step 202, performing correlation analysis on the generated sample sets to generate corresponding classification combinations.

Step 203: Determine the conversation topic corresponding to the classification combination according to the occurrence frequency of words in each classification combination, and associate the conversation topic with the conversation record corresponding to the classification combination.

Step 204 , analyzing the correlation between the conversation topics, and merging the conversation topics whose correlation is greater than a predetermined threshold into a same conversation topic, so that the merged conversation topic is associated with the conversation records corresponding to all the merged conversation topics.

Step 205, when the user searches for information in the conversation record, search for a conversation topic matching the keyword according to the keyword entered by the user when querying, and present the conversation record associated with the found conversation topic to the user.

In step 201, the processing flow for classifying session records is shown in Figure 3, and the processing is as follows:

Step 301 , judging whether the session record has been classified, and if it has been classified, it is not processed; otherwise, step 302 is executed.

Step 302, classify the session records that have not been classified according to different users, such as: judge whether the session records TRi and the session records TR _j belong to the session records between the same users, if the session records TRi and the session records TR _j Belonging to sessions between different users, the session records TRi and session records TR _j are divided into different sample sets TS; if the session records TRi and session records TR _j belong to the session records between the same users, the session records TRi and session records TR _j is divided into the same sample set.

Step 303 : Divide the same sample set according to the interval time of conversation records in the sample set, and further divide it into different sample sets. The interval time of conversation records can be set to one week according to the actual application.

The sample set TS generated through step 303 is a sample set for correlation analysis.

Referring to Figure 4, the process of performing correlation analysis on a sample set using the KNN (K Nearest Neighbor, K nearest neighbor) algorithm is as follows:

(W₁，W₂，W₃......Wn)，其中Wi为第i个元素的权重，各元素为S中的词语。Step 401. Generate a corresponding feature vector for each session record TR in the sample set TS. First, perform word segmentation processing on each session record TR, remove the auxiliary words, interjections and other meaningless words, and obtain the set S; merge the synonyms in S, for example, merge {"computer", "computer"} into {"computer","computer"}. Vectorize the set S corresponding to each session record after merging synonyms to generate a feature vector

(W ₁ , W ₂ , W ₃ ......Wn), where Wi is the weight of the i-th element, and each element is a word in S.

中各元素的权值W。采用如下公式进行权值计算：Step 402, calculating feature vectors corresponding to each session record TR

The weight W of each element in . Use the following formula to calculate the weight:

$\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; tf</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.01</span>}<span\; class="notranslate">\; )</span>}{\sqrt{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Element;</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; tf</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; N</span>}{<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.01</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

为词t在特征向量

中的权重，而为词t在特征向量中的词频，N为每个样本集合TS中会话记录TR的总数，n_t为每个样本集合TS中出现词t的会话记录TR数，分母为归一化因子。in,

is the word t in the feature vector

weights in , while is the word t in the feature vector The word frequency in , N is the total number of conversation records TR in each sample set TS, n _t is the number of conversation record TRs where word t appears in each sample set TS, and the denominator is the normalization factor.

Step 403: Calculate correlation coefficients between feature vectors corresponding to each session record, and determine K feature vectors most similar to each feature vector according to the calculated correlation coefficients.

For specific implementation, the following formula is used $\mathrm{Sim}(d_i,d_j)=\frac{\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ik}}\&times;W_{\mathrm{jk}}}{\sqrt{\left(\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ik}}^2\right)\left(\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{W}_{\mathrm{jk}}^2\right)}}$ Calculate the correlation coefficient between the eigenvectors corresponding to each session record, where Sim(d _i , d _j ) is the correlation coefficient between the eigenvector d _i and the eigenvector d _j , W _ik and W _jk are the eigenvectors d _i and The weight of the kth element of the feature vector _dj .

Through calculation, the correlation coefficient between the feature vectors is obtained, and according to the correlation coefficient, the K feature vectors most correlated with each feature vector are combined into a set, and the value of K can be determined according to the actual application.

Step 404: Divide the feature vectors corresponding to each session record into different classes in the classification C to generate a classification combination.

Classification C is a set composed of feature vectors corresponding to each session record in the sample set TS.

Method 1: When category C is empty, generate a vector set c in category C in the following way, and then add c to category C:

和特征向量

分别属于对方最相似的K个邻居组成的集合，则

和

属于同一类c，生成类c并将该类与特征向量

和特征向量对应的会话记录关联，然后将类c添加到分类C，每个类c中的特征向量组成一个分类组合。A feature vector corresponding to the session record

and eigenvectors

belong to the set of K neighbors most similar to each other, then

and

belong to the same class c, generate class c and combine this class with the feature vector

and eigenvectors The corresponding session records are associated, and then class c is added to classification C, and the feature vectors in each class c form a classification combination.

属于某个类c(c∈C)的权重，采用如下的公式：Method 2: When the category C is not empty, calculate the feature vector corresponding to each session record

The weight belonging to a certain class c (c∈C) adopts the following formula:

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; KNN</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; Sim</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; ,</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; d</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

为对应于一条会话记录的特征向量，

为与

最相似的K个邻居组成的集合中的特征向量，

为

与其最相关的特征向量

的相关系数，该相关系数可根据步骤403计算结果获得，

为类别属性函数，如果特征向量

属于类C_j，

的函数值为1，否则为0。根据计算得到比较特征向量

在各类C_j中的权值，将特征向量

分到权值较大的类C_j中，并将该类C_j与特征向量

对应的会话记录关联。in,

is the feature vector corresponding to a session record,

for with

The eigenvectors in the set of the most similar K neighbors,

for

The eigenvector most correlated with it

The correlation coefficient, which can be obtained according to the calculation result of step 403,

is a category attribute function, if the feature vector

belongs to class C _j ,

The value of the function is 1, otherwise it is 0. According to the calculation compare eigenvectors

The weights in each category C _j , the eigenvector

Classify into the class C _j with larger weight, and combine the class C _j with the feature vector

Corresponding session record association.

和现存每个类c的相关度都很小，则可采用方法一的方式生成一个新的类c′，并将类c′加入到分类C中，并将类c′与特性向量

对应的会话记录关联。When using the second method, if the eigenvector

The correlation with each existing class c is very small, then a new class c' can be generated by method 1, and class c' is added to the classification C, and the class c' and the feature vector

Corresponding session record association.

After processing each eigenvector, the eigenvectors are all divided into one class, and each class forms a classification combination.

The N words with the highest frequency of occurrence in each generated classification combination or the words with a frequency greater than a are determined as the conversation topic of the classification combination, and the N value and a value are determined according to the actual application.

After performing the above processing on each sample set TS, a classification combination and a conversation topic corresponding to the classification combination are generated. When performing correlation analysis on the generated conversation topic, the conversation topic is regarded as a sample collection of the KNN algorithm, and each of the conversation topics in the collection is calculated. According to the weight of each word in the conversation topic in the conversation topic, the formula in step 403 is used to calculate the correlation coefficient of each conversation topic, and the conversation topics whose correlation coefficient is greater than the set threshold are merged.

When presenting the session records to the user, the session records are arranged according to different session users, and can also be arranged according to the weight order of the session records in the session topic.

In the above embodiments, the KNN algorithm is used to analyze the correlation of the sample set, but the present invention is not limited to using the KNN algorithm to analyze the sample set. The method for analyzing the correlation of session records can also apply vector space-based training algorithms and classification methods such as vector machine algorithms, neural network algorithms, and Bayesian algorithms. For example, when using the Bayesian algorithm, calculate the probability that each word in the corresponding feature vector of each session record appears in a certain session, and then calculate the probability that the feature vector belongs to a certain session according to the Bayesian formula, and add it to the probability largest session.

With the present invention, when the user inquires the information that the user cares about in the conversation records, the user only needs to input keywords, and the system will automatically search for the conversation topics that match the keywords, and present the conversation records associated with the conversation topics to the user, It not only avoids the cumbersome operation when users manually query information, but also improves the query efficiency.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (8)

1. A management method for instant communication session records is characterized by comprising the following steps:

acquiring session records of a user and classifying the session records to obtain a sample set;

generating a feature vector corresponding to each session record in the sample set, analyzing the correlation between each feature vector and other feature vectors, and classifying the feature vectors according to the correlation to generate a classification combination;

determining conversation topics corresponding to the classification combinations according to the occurrence frequency of the words in each classification combination, and enabling the conversation topics to be associated with conversation records corresponding to the classification combinations; and

and searching a conversation theme matched with the keyword according to the keyword input by the user during the query, and presenting the searched conversation record associated with the conversation theme to the user.

2. The method of claim 1, wherein after the conversation topics are generated, the correlation between the conversation topics is further analyzed, and the conversation topics with the correlation larger than a predetermined threshold are combined into the same conversation topic, so that the combined conversation topic is associated with the conversation records corresponding to all the combined conversation topics.

3. The method of claim 1 or 2, wherein the set of samples is generated by classifying session records by different session users.

4. The method of claim 3, wherein a sample set is further divided into a plurality of different sample sets according to the time intervals of session records in the sample set.

5. The method according to claim 1, wherein the generating a feature vector corresponding to each session record in the sample set, and analyzing the correlation between each feature vector and other feature vectors specifically includes:

performing word segmentation processing on each conversation record, deleting words without practical meaning in the conversation record to obtain a set S, merging synonyms in the set S, and performing vectorization to generate a feature vector corresponding to the conversation record

(W₁，W₂，W₃.... Wn), where Wi is the weight of the ith element, each element being a word in S;

computing correspondence to session recordsFeature vector of

And calculating the relevance of each feature vector according to the weight of each word forming the feature vector in the feature vector.

6. The method of claim 1, wherein the topic of conversation for a taxonomic group is determined according to terms in the taxonomic group that occur more frequently than a predetermined threshold.

7. An apparatus for managing instant messaging session records, comprising:

means for storing a user session record;

means for classifying the session records to generate a sample set;

a unit for generating a feature vector corresponding to each session record in the sample set, analyzing the correlation between each feature vector and other feature vectors, and classifying the feature vectors according to the correlation to generate a classification combination;

a unit for determining a conversation topic corresponding to the classification combination and associating the conversation topic to a conversation record corresponding to the classification combination; and

and the unit is used for searching the conversation theme matched with the keyword according to the keyword input by the user during the query and presenting the searched conversation record associated with the conversation theme to the user.

8. The apparatus of claim 7, further comprising:

and the unit is used for analyzing the correlation among the conversation topics, combining the conversation topics with the correlation larger than a preset threshold value into the same conversation topic, and associating the combined conversation topic with the conversation records corresponding to all the combined conversation topics.

Images