User grouping method, device, equipment and storage medium

Abstract

The present invention relates to the field of big data, and discloses a user grouping method, device, equipment and storage medium. The method comprises: obtaining human-computer interaction voice data of each user in multiple business links, and preprocessing the human-computer interaction voice data to obtain user intention data; extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding the user intentions to obtain a user intention matrix; using a preset matrix decomposition method to perform matrix decomposition on the user intention matrix to obtain an intention feature matrix; according to the intention feature matrix, grouping the users to obtain multiple user groups. The present invention greatly improves the accuracy of user grouping by analyzing user intentions.

Description

Technical Field

[0001] The present invention relates to the field of big data, and in particular to a user grouping method, device, equipment and storage medium. Background Art

[0002] With the rapid development of computer technology, using robots to communicate with users and handle user needs is very efficient and can save a lot of labor costs. Users can show their behavioral preferences through the operation of mobile terminals. By dividing user groups according to their behavioral preferences, pushing similar information to users in the same group, or using similar users to push information to target users, personalized content can be pushed to users accurately.

[0003] Existing user grouping methods mostly group users based on their behavioral characteristics and interests. This method only takes into account the characteristics of the user, resulting in the grouping results being only applicable to specific application scenarios and having poor portability. Therefore, the accuracy of existing user grouping technologies is low and unsatisfactory. Summary of the invention

[0004] The main purpose of the present invention is to solve the current technical problem of low accuracy of user grouping.

[0005] A first aspect of the present invention provides a user grouping method, comprising:

[0006] Acquire human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0007] Extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding each user intention to obtain a user intention matrix;

[0008] Performing matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0009] According to the intention feature matrix, the users are grouped to obtain multiple user groups.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the acquiring of human-computer interaction voice data of each user in multiple business links, and preprocessing of the human-computer interaction voice data to obtain user intention data includes:

[0011] Obtain human-computer interaction voice data of each user in multiple business links;

[0012] Performing speech recognition and text conversion on the human-computer interaction speech data to obtain speech-to-text data;

[0013] The speech text data is semantically analyzed and classified to obtain user intention data.

[0014] Optionally, in a second implementation of the first aspect of the present invention, extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding each user intention to obtain a user intention matrix includes:

[0015] Extracting vocabulary corresponding to each human-computer interaction skill point in the user intention data;

[0016] Performing word vector conversion on the words respectively to obtain multiple word vectors;

[0017] Based on a preset user intent information library, context analysis is performed on the word vectors to obtain multiple user intents;

[0018] The user intentions are encoded to obtain a user intention matrix.

[0019] Optionally, in a third implementation manner of the first aspect of the present invention, performing matrix decomposition on the user intention matrix to obtain an intention feature matrix includes:

[0020] Calculating a basis matrix of the user intention matrix, and calculating a coefficient matrix of the user intention matrix based on the basis matrix;

[0021] The base matrix and the coefficient matrix are alternately iterated to obtain an intention feature matrix.

[0022] Optionally, in a fourth implementation manner of the first aspect of the present invention, the alternately iterating the basis matrix and the coefficient matrix to obtain the intention feature matrix includes:

[0023] Initializing the base matrix and the coefficient matrix respectively to obtain an initial base matrix and an initial coefficient matrix;

[0024] Iteratively calculating the initial basis matrix and the initial coefficient matrix to obtain an updated basis matrix;

[0025] Call the preset loss function to calculate the loss value of the current updated basis matrix and the initial basis matrix until the loss value is less than the preset loss threshold, and then use the current updated basis matrix as the intention feature matrix.

[0026] Optionally, in a fifth implementation of the first aspect of the present invention, grouping the users according to the intention feature matrix to obtain a plurality of user groups includes:

[0027] Extracting column vectors from the intention feature matrix to obtain multiple column vectors;

[0028] Performing eigenvalue calculations on each of the column vectors respectively to obtain a plurality of eigenvalues ​​corresponding to each of the column vectors;

[0029] The maximum values ​​of the multiple eigenvalues ​​corresponding to the column vectors are respectively extracted, and the maximum values ​​are used as valid user groups of the corresponding column vectors to obtain multiple user groups.

[0030] Optionally, in a sixth implementation manner of the first aspect of the present invention, after grouping the users according to the intention feature matrix to obtain a plurality of user groups, the method further includes:

[0031] Labeling each user to obtain multiple user labels;

[0032] Clustering the user tags according to the user groups to obtain user tag sets corresponding to the user groups;

[0033] The user feature distribution is drawn according to the user tag sets corresponding to the user groups to obtain the user group portraits.

[0034] A second aspect of the present invention provides a user grouping device, the user grouping device comprising:

[0035] An acquisition module is used to acquire human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0036] An encoding module is used to extract the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encode the user intentions to obtain a user intention matrix;

[0037] A matrix decomposition module, used to perform matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0038] The user grouping module is used to group the users according to the intention feature matrix to obtain multiple user groups.

[0039] Optionally, in a first implementation of the second aspect of the present invention, the acquisition module is specifically used to:

[0040] Obtain human-computer interaction voice data of each user in multiple business links;

[0041] Performing speech recognition and text conversion on the human-computer interaction speech data to obtain speech-to-text data;

[0042] The speech text data is semantically analyzed and classified to obtain user intention data.

[0043] Optionally, in a second implementation of the second aspect of the present invention, the encoding module is specifically used to:

[0044] Extracting vocabulary corresponding to each human-computer interaction skill point in the user intention data;

[0045] Performing word vector conversion on the words respectively to obtain multiple word vectors;

[0046] Based on a preset user intent information library, context analysis is performed on the word vectors to obtain multiple user intents;

[0047] The user intentions are encoded to obtain a user intention matrix.

[0048] Optionally, in a third implementation of the second aspect of the present invention, the matrix decomposition module further includes:

[0049] A matrix calculation unit, used to calculate a basis matrix of the user intention matrix, and calculate a coefficient matrix of the user intention matrix based on the basis matrix;

[0050] The alternating calculation unit is used to perform alternating iterations on the base matrix and the coefficient matrix to obtain an intention feature matrix.

[0051] Optionally, in a fourth implementation manner of the second aspect of the present invention, the alternating calculation unit is specifically used to:

[0052] Initializing the base matrix and the coefficient matrix respectively to obtain an initial base matrix and an initial coefficient matrix;

[0053] Iteratively calculating the initial basis matrix and the initial coefficient matrix to obtain an updated basis matrix;

[0054] Call the preset loss function to calculate the loss value of the current updated basis matrix and the initial basis matrix until the loss value is less than the preset loss threshold, and then use the current updated basis matrix as the intention feature matrix.

[0055] Optionally, in a fifth implementation of the second aspect of the present invention, the user grouping module is specifically used to:

[0056] Extracting column vectors from the intention feature matrix to obtain multiple column vectors;

[0057] Performing eigenvalue calculations on each of the column vectors respectively to obtain a plurality of eigenvalues ​​corresponding to each of the column vectors;

[0058] The maximum values ​​of the multiple eigenvalues ​​corresponding to the column vectors are respectively extracted, and the maximum values ​​are used as valid user groups of the corresponding column vectors to obtain multiple user groups.

[0059] Optionally, in a sixth implementation of the second aspect of the present invention, the user grouping device further includes:

[0060] The portrait generation module is used to label each user to obtain multiple user labels; cluster each user label according to each user group to obtain a user label set corresponding to each user group; draw a user feature distribution according to the user label set corresponding to each user group to obtain a user group portrait.

[0061] The third aspect of the present invention provides a user grouping device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the user grouping device executes the above-mentioned user grouping method.

[0062] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned user grouping method.

[0063] In the technical solution provided by the present invention, by obtaining the human-computer interaction voice data of each user in multiple business links, the obtained human-computer interaction voice data is preprocessed, and the purpose of the preprocessing is to extract the user intention data in the voice data; extract the user intention corresponding to the human-computer interaction skill point pre-set in the business link, obtain multiple user intentions, encode the user intentions, and obtain a user intention matrix containing multiple users and multiple user intentions; perform matrix decomposition on the user intention matrix to obtain an intention feature matrix; according to the intention feature matrix, the users are grouped to obtain multiple user groups, because the intention feature matrix can represent the common behavior pattern composed of multiple different user demographic data, so the user group with the largest coefficient in each column of the matrix is ​​taken as the effective user group. The present invention improves the accuracy of user grouping, and the needs of different categories of users can be better analyzed through more accurate user groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a schematic diagram of a first embodiment of a method for grouping users in an embodiment of the present invention;

[0065] Figure 2 Schematic diagram of a second embodiment of the user grouping method in an embodiment of the present invention;

[0066] Figure 3 Schematic diagram of a third embodiment of the user grouping method in an embodiment of the present invention;

[0067] Figure 4 Schematic diagram of a fourth embodiment of the user grouping method according to an embodiment of the present invention;

[0068] Figure 5 A schematic diagram of an embodiment of a user grouping device in an embodiment of the present invention;

[0069] Figure 6 The figure is a schematic diagram of an embodiment of a user grouping device in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] Embodiments of the present invention provide a user grouping method, apparatus, device and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0071] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , a first embodiment of the user grouping method in the embodiment of the present invention includes:

[0072] 101. Obtain human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0073] It is understandable that the execution subject of the present invention may be a user grouping device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0074] In this embodiment, ASR technology and NLP technology are used to pre-process the human-computer interaction voice data. Among them, ASR technology performs voice recognition and text conversion on the human-computer interaction voice data, and NLP technology performs semantic analysis and classification on the voice data after ASR processing, and finally obtains user intention data.

[0075] Optionally, in one embodiment, the above step 101 includes:

[0076] Obtain human-computer interaction voice data of each user in multiple business links;

[0077] Performing speech recognition and text conversion on the human-computer interaction speech data to obtain speech-to-text data;

[0078] The speech text data is semantically analyzed and classified to obtain user intention data.

[0079] In this embodiment, the process of speech recognition is to cut the sound signal into small segments, and then each segment is represented by a vector according to certain rules. The compiled vector is put into the acoustic model to get the letter corresponding to each segment. Then the translated letters are passed through the language model to assemble the speech text data. Semantic analysis is to cut a sentence into multiple words, classify the themes of these words, such as economics, sports, literature, etc., to obtain the user's intention data.

[0080] 102. Extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding the user intentions to obtain a user intention matrix;

[0081] In this embodiment, the human-machine interaction skill point refers to a certain ability of the robot, such as weather checking skill and ticket booking skill. The user intentions are encoded to obtain a user intention matrix.

[0082] For example: g users, use G1, G2, ...G g To represent, each user has c intention sequences C1, C2, ...C c , let V represent the matrix of g*c, and the single element V of the matrix ij The value represents the G i User's C j intention.

[0083] 103. Perform matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0084] In this embodiment, the matrix decomposition method adopts the NMF algorithm. NMF is to find the non-negative matrix W and the non-negative matrix H given the matrix V, so that V = W*H. It can be understood that the column vector of the original matrix is ​​the weighted sum of all column vectors in the left matrix, and the weight coefficient is the element of the corresponding column vector of the right matrix, so W is called the base matrix, H is the coefficient matrix, and the base matrix W is the intention feature matrix. By using the base matrix instead of the original matrix, the original matrix can be reduced in dimension to obtain the reduced-dimensional matrix of the data features, thereby reducing the storage space.

[0085] 104. Group the users according to the intention feature matrix to obtain multiple user groups.

[0086] In this embodiment, the matrix V is decomposed to obtain a matrix H (containing user intent) and another matrix W (containing demographic data). W represents a common behavior pattern composed of multiple different user demographic data, and the user with the largest coefficient in each column can be regarded as the most effective user group.

[0087] Optionally, in one embodiment, after the above step 104, the following further steps are included:

[0088] Labeling each user to obtain multiple user labels;

[0089] In this embodiment, the user portrait is a labeled user model abstracted from the user's data, that is, user data labeling, that is, refining the data and labeling it.

[0090] Clustering the user tags according to the user groups to obtain user tag sets corresponding to the user groups;

[0091] The user feature distribution is drawn according to the user tag sets corresponding to the user groups to obtain the user group portraits.

[0092] In this embodiment, data is sorted according to the user intentions of each group in the user clustering group, the user tags are clustered to obtain a user tag set corresponding to each user group, and a user portrait for each user clustering is constructed. The user portrait is enriched by demographic data and telemarketing intentions to provide a reference for the construction of sales talk. At the same time, the user portraits of different categories of users can better analyze the needs of different categories of users and provide exclusive services for different categories of users.

[0093] In an embodiment of the present invention, by obtaining human-computer interaction voice data of each user in multiple business links, the obtained human-computer interaction voice data is preprocessed, and the purpose of the preprocessing is to extract user intention data in the voice data; extract the user intention corresponding to the human-computer interaction skill point pre-set in the business link to obtain multiple user intentions, encode the user intentions, and obtain a user intention matrix containing multiple users and multiple user intentions; perform matrix decomposition on the user intention matrix to obtain an intention feature matrix; according to the intention feature matrix, group the users to obtain multiple user groups, because the intention feature matrix can represent the common behavior pattern composed of multiple different user demographic data, so the user group with the largest coefficient in each column of the matrix is ​​taken as the effective user group. The present invention improves the accuracy of user grouping, and the needs of different categories of users can be better analyzed through more accurate user groups.

[0094] See also Figure 2 The second embodiment of the user grouping method in the embodiment of the present invention includes:

[0095] 201. Obtaining human-computer interaction voice data of each user in multiple business links, and preprocessing the human-computer interaction voice data to obtain user intention data;

[0096] 202. Extracting vocabulary corresponding to each human-computer interaction skill point in the user intention data;

[0097] In this embodiment, the vocabulary corresponding to the TFIDF human-computer interaction skill points and other means based on word frequency statistics are used, which are usually used for the vector representation of sentences.

[0098] For example: Help me book a restaurant for the evening, then the vocabulary corresponding to the human-computer interaction skill point is "restaurant reservation".

[0099] For example: What will the weather be like tomorrow? The corresponding word for the human-computer interaction skill point is "weather query".

[0100] 203. Perform word vector conversion on the words respectively to obtain multiple word vectors;

[0101] In this embodiment, the words in the training set are mapped into a high-dimensional space. In this space, the distance between words can reflect the similarity of the words themselves, and the meaning can be converted by adding and subtracting the word vectors themselves.

[0102] 204. Based on a preset user intention information database, context analysis is performed on the word vectors to obtain multiple user intentions;

[0103] In this embodiment, when a user queries a target word, the real user semantics is identified by performing context analysis on the word vector, and then the real user intention is discovered through identification through the user intention information library.

[0104] 205. Encode each user intention to obtain a user intention matrix;

[0105] In this embodiment, the intention of each user is encoded. For example, if there are g users, G1, G2, ... G g To represent, each user has c intention sequences C1, C2, ...C c , let V represent the matrix of g*c, and the single element D of the matrix, D ij The value represents the G i User's C j intention.

[0106] 206. Perform matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0107] 207. Group the users according to the intention feature matrix to obtain multiple user groups.

[0108] In an embodiment of the present invention, a matrix containing user intentions is obtained by analyzing user intention data. The construction of the user intention matrix can associate users with the intentions corresponding to each user, so that multiple users and the intentions of multiple users are collected into one matrix, and a certain element in the matrix can represent a certain intention of the user, so that user grouping can obtain groups according to user intentions.

[0109] See also Figure 3 A third embodiment of the user grouping method in the embodiment of the present invention includes:

[0110] 301. Obtain human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0111] 302. Extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding the user intentions to obtain a user intention matrix;

[0112] 303. Calculate a basis matrix of the user intention matrix, and calculate a coefficient matrix of the user intention matrix according to the basis matrix;

[0113] 304. Alternately iterate the base matrix and the coefficient matrix to obtain an intention feature matrix;

[0114] In this embodiment, the NMF algorithm can find a non-negative matrix H and a non-negative matrix W, decompose a non-negative matrix into the product of two non-negative matrices on the left and right, reduce the dimension of the matrix, and compress a large amount of data. NMF is an unsupervised learning algorithm, in which the restriction condition is that all elements in W and H must be greater than 0. Since the matrices before and after decomposition contain only non-negative elements, a column vector in the original matrix V (such as the dotted column vector in V in the above figure) can be interpreted as the weighted sum of all column vectors (basis vectors) in the basis matrix W, and the weight coefficient is the element (weight) in the corresponding column vector in the coefficient matrix H.

[0115] Optionally, in one embodiment, the above step 304 includes:

[0116] Initializing the base matrix and the coefficient matrix respectively to obtain an initial base matrix and an initial coefficient matrix;

[0117] Iteratively calculating the initial basis matrix and the initial coefficient matrix to obtain an updated basis matrix;

[0118] Call the preset loss function to calculate the loss value of the current updated basis matrix and the initial basis matrix until the loss value is less than the preset loss threshold, and then use the current updated basis matrix as the intention feature matrix.

[0119] In this embodiment, the NMF algorithm calculates the basis matrix W of the intention feature matrix V, and then fixes the basis matrix W to solve the coefficient matrix H, and initializes the basis matrix W and the coefficient matrix H. The initialization is to fill the basis matrix and the coefficient matrix with non-negative numbers to obtain the initial basis matrix and the initial coefficient matrix. The iterative calculation process is to fix the basis matrix to calculate the coefficient matrix, and then fix the coefficient matrix to iterate. The loss value of the updated basis matrix after iteration and the initial basis matrix is ​​calculated to determine whether the NMF algorithm converges. When the loss value is less than the loss threshold, the current updated basis matrix is ​​used as the intention feature matrix. This embodiment uses the NMF algorithm to find the basis matrix W and the coefficient matrix H, so that the intention feature matrix V = W*H. Among them, the i-th column of V is equal to W multiplied by the i-th column of H, that is, the j-th element of the i-th column of H is equivalent to the weight of the j-th column of W, and the i-th column of V is the sum of the products of each column of W and the weight. In other words, each column of W is equivalent to a basis vector, and each column of H is equivalent to a coordinate vector.

[0120] 305. Group the users according to the intention feature matrix to obtain multiple user groups.

[0121] In the embodiment of the present invention, the intent feature matrix is ​​decomposed by the NMF algorithm, and the original problem of clustering the intent feature matrix is ​​converted into the problem of clustering the base matrix through dimensionality reduction. The NMF algorithm can achieve the effect of representing the correlation between the local parts of the user's intention, thereby obtaining better user clustering results. Matrix decomposition technology accelerates the clustering of user intentions and improves the efficiency of user clustering by decomposing the user intent matrix.

[0122] See also Figure 4 , a fourth embodiment of the user grouping method in the embodiment of the present invention includes:

[0123] 401. Obtain human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0124] 402. Extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding the user intentions to obtain a user intention matrix;

[0125] 403. Perform matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0126] 404. Extracting column vectors from the intention feature matrix to obtain multiple column vectors;

[0127] In this embodiment, the column vectors of the intention feature matrix represent multiple users with a common intention.

[0128] 405. Perform eigenvalue calculation on each column vector respectively to obtain multiple eigenvalues ​​corresponding to each column vector;

[0129] In this embodiment, the characteristic value is calculated using the following formula:

[0130]

[0131] Among them, u jk represents the eigenvalue corresponding to the column vector, v i For the i-th user, v j represents the jth intention, c represents the cth intention, x k represents a custom parameter, and x k ≠ v j , m is a positive integer greater than 1.

[0132] 406. Extract the maximum values ​​of the multiple eigenvalues ​​corresponding to the column vectors respectively, and use the maximum values ​​as valid user groups of the corresponding column vectors to obtain multiple user groups.

[0133] In this embodiment, the multiple eigenvalues ​​corresponding to the column vectors are arranged in order from small to large, and the maximum value is selected, which can be used as an effective user group. Each column vector corresponds to a user group, and multiple user groups can be obtained.

[0134] In an embodiment of the present invention, the eigenvalue of each column vector is calculated by the eigenvalue formula to obtain the eigenvalue corresponding to each element in each column vector, and then the eigenvalues ​​of each column vector are compared respectively. The one with the largest eigenvalue is taken as the most effective user grouping. By analyzing the column vectors of the intention feature matrix, user grouping is made more accurate.

[0135] The above describes the user grouping method in the embodiment of the present invention. The following describes the user grouping device in the embodiment of the present invention. Figure 5 , an embodiment of the user grouping device in the embodiment of the present invention includes:

[0136] The acquisition module 501 is used to acquire the human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data;

[0137] The encoding module 502 is used to extract the user intention corresponding to each human-computer interaction skill point in the user intention data, obtain multiple user intentions, and encode the user intentions to obtain a user intention matrix;

[0138] A matrix decomposition module 503 is used to perform matrix decomposition on the user intention matrix to obtain an intention feature matrix;

[0139] The user grouping module 504 is used to group the users according to the intention feature matrix to obtain multiple user groups.

[0140] Optionally, in one embodiment, the acquisition module 501 is specifically used for:

[0141] Obtain human-computer interaction voice data of each user in multiple business links;

[0142] Performing speech recognition and text conversion on the human-computer interaction speech data to obtain speech-to-text data;

[0143] The speech text data is semantically analyzed and classified to obtain user intention data.

[0144] Optionally, in one embodiment, the encoding module 502 is specifically used for:

[0145] Extracting vocabulary corresponding to each human-computer interaction skill point in the user intention data;

[0146] Performing word vector conversion on the words respectively to obtain multiple word vectors;

[0147] Based on a preset user intent information library, context analysis is performed on the word vectors to obtain multiple user intents;

[0148] The user intentions are encoded to obtain a user intention matrix.

[0149] Optionally, in one embodiment, the matrix decomposition module 503 further includes:

[0150] A matrix calculation unit, used to calculate a basis matrix of the user intention matrix, and calculate a coefficient matrix of the user intention matrix based on the basis matrix;

[0151] The alternating calculation unit is used to perform alternating iterations on the base matrix and the coefficient matrix to obtain an intention feature matrix.

[0152] Optionally, in one embodiment, the alternating calculation unit is specifically used for:

[0153] Initializing the base matrix and the coefficient matrix respectively to obtain an initial base matrix and an initial coefficient matrix;

[0154] Iteratively calculating the initial basis matrix and the initial coefficient matrix to obtain an updated basis matrix;

[0155] Call the preset loss function to calculate the loss value of the current updated basis matrix and the initial basis matrix until the loss value is less than the preset loss threshold, and then use the current updated basis matrix as the intention feature matrix.

[0156] Optionally, in one embodiment, the user grouping module 504 is specifically used for:

[0157] Extracting column vectors from the intention feature matrix to obtain multiple column vectors;

[0158] Performing eigenvalue calculations on each of the column vectors respectively to obtain a plurality of eigenvalues ​​corresponding to each of the column vectors;

[0159] The maximum values ​​of the multiple eigenvalues ​​corresponding to the column vectors are respectively extracted, and the maximum values ​​are used as valid user groups of the corresponding column vectors to obtain multiple user groups.

[0160] Optionally, in one embodiment, the user grouping device further includes:

[0161] The portrait generation module 505 is used to label each user to obtain multiple user labels; cluster each user label according to each user group to obtain a user label set corresponding to each user group; draw a user feature distribution according to the user label set corresponding to each user group to obtain a user group portrait.

[0162] In an embodiment of the present invention, by obtaining human-computer interaction voice data of each user in multiple business links, the obtained human-computer interaction voice data is preprocessed, and the purpose of the preprocessing is to extract user intention data in the voice data; extract the user intention corresponding to the human-computer interaction skill point pre-set in the business link to obtain multiple user intentions, encode the user intentions, and obtain a user intention matrix containing multiple users and multiple user intentions; perform matrix decomposition on the user intention matrix to obtain an intention feature matrix; according to the intention feature matrix, group the users to obtain multiple user groups, because the intention feature matrix can represent the common behavior pattern composed of multiple different user demographic data, so the user group with the largest coefficient in each column of the matrix is ​​taken as the effective user group. The present invention improves the accuracy of user grouping, and the needs of different categories of users can be better analyzed through more accurate user groups.

[0163] above Figure 5The user grouping apparatus in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The user grouping device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0164] Figure 6 : is a structural diagram of a user grouping device provided by an embodiment of the present invention. The user grouping device 600 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (for example, one or more processors) and a memory 620, and one or more storage media 630 (for example, one or more mass storage devices) storing application programs 633 or data 632. Among them, the memory 620 and the storage medium 630 can be short-term storage or permanent storage. The program stored in the storage medium 630 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the user grouping device 600. Furthermore, the processor 610 can be configured to communicate with the storage medium 630 to execute a series of instruction operations in the storage medium 630 on the user grouping device 600.

[0165] The user grouping device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and / or one or more operating systems 631, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 6 The structure of the user grouping device shown does not constitute a limitation on the user grouping device, and may include more or less components than those shown in the figure, or combine certain components, or arrange the components differently.

[0166] The present invention also provides a user grouping device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the user grouping method in the above embodiments.

[0167] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the user grouping method.

[0168] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0169] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0170] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A user grouping method, characterized in that: The user grouping method comprises: Acquire human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data; Extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding each user intention to obtain a user intention matrix; Performing matrix decomposition on the user intention matrix to obtain an intention feature matrix; According to the intention feature matrix, the users are grouped to obtain a plurality of user groups; The grouping of the users according to the intention feature matrix to obtain a plurality of user groups includes: Extracting column vectors from the intention feature matrix to obtain multiple column vectors; The eigenvalues ​​of the column vectors are calculated respectively to obtain multiple eigenvalues ​​corresponding to the column vectors. The calculation formulas corresponding to the multiple eigenvalues ​​are as follows: Among them, u jk represents multiple eigenvalues ​​corresponding to each column vector, v i For the i-th user, v j represents the jth intention, c represents the cth intention, x k represents a custom parameter, and x k ≠v j , m is a positive integer greater than 1, i, j and k are all positive integers; The maximum values ​​of the multiple eigenvalues ​​corresponding to the column vectors are respectively extracted, and the maximum values ​​are used as valid user groups of the corresponding column vectors to obtain multiple user groups.

2. The user grouping method according to claim 1, characterized in that: The acquiring of human-computer interaction voice data of each user in multiple business links and preprocessing of the human-computer interaction voice data to obtain user intention data includes: Obtain human-computer interaction voice data of each user in multiple business links; Performing speech recognition and text conversion on the human-computer interaction speech data to obtain speech-to-text data; The speech text data is semantically analyzed and classified to obtain user intention data.

3. The user grouping method according to claim 2, characterized in that: The extracting the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encoding each user intention to obtain a user intention matrix includes: Extracting vocabulary corresponding to each human-computer interaction skill point in the user intention data; Performing word vector conversion on the words respectively to obtain multiple word vectors; Based on a preset user intent information library, context analysis is performed on the word vectors to obtain multiple user intents; The user intentions are encoded to obtain a user intention matrix.

4. The user grouping method according to claim 1, characterized in that: The performing matrix decomposition on the user intention matrix to obtain the intention feature matrix comprises: Calculating a basis matrix of the user intention matrix, and calculating a coefficient matrix of the user intention matrix based on the basis matrix; The base matrix and the coefficient matrix are alternately iterated to obtain an intention feature matrix.

5. The user grouping method according to claim 4, characterized in that: The alternately iterating the base matrix and the coefficient matrix to obtain the intention feature matrix comprises: Initializing the base matrix and the coefficient matrix respectively to obtain an initial base matrix and an initial coefficient matrix; Iteratively calculating the initial basis matrix and the initial coefficient matrix to obtain an updated basis matrix; Call the preset loss function to calculate the loss value of the current updated basis matrix and the initial basis matrix until the loss value is less than the preset loss threshold, and then use the current updated basis matrix as the intention feature matrix.

6. The user grouping method according to any one of claims 1 to 5, characterized in that: After grouping the users according to the intention feature matrix to obtain a plurality of user groups, the method further includes: Labeling each user to obtain multiple user labels; Clustering the user tags according to the user groups to obtain user tag sets corresponding to the user groups; The user feature distribution is drawn according to the user tag sets corresponding to the user groups to obtain the user group portraits.

7. A user grouping device, executing the user grouping method according to any one of claims 1 to 6, characterized in that: The user grouping device comprises: An acquisition module is used to acquire human-computer interaction voice data of each user in multiple business links, and pre-process the human-computer interaction voice data to obtain user intention data; An encoding module is used to extract the user intention corresponding to each human-computer interaction skill point in the user intention data to obtain multiple user intentions, and encode the user intentions to obtain a user intention matrix; A matrix decomposition module, used to perform matrix decomposition on the user intention matrix using a preset matrix decomposition method to obtain an intention feature matrix; The user grouping module is used to group the users according to the intention feature matrix to obtain multiple user groups.

8. A user grouping device, characterized in that: The user grouping device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the user grouping device to execute the user grouping method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the user grouping method according to any one of claims 1 to 6 is implemented.

Images