Training methods for psychological profile recognition models and collection action recommendation systems

Abstract

The present invention discloses a training method for a psychological portrait recognition model, which obtains the score values ​​of N sample objects on J psychological evaluation labels as samples to train the psychological portrait recognition model, maps the scores of J different evaluation dimensions into Euclidean space in the form of coordinates, and assigns different objects to appropriate psychological portrait categories in combination with cluster analysis. The psychological portrait recognition process can achieve automatic learning and optimization, thereby improving the accuracy of classification. The present invention also provides a method and system for recommending collection actions using the psychological portrait recognition model obtained through the above training, which provides collection action recommendations that match the psychological characteristics of the collection object by identifying the psychological portrait category of the collection object, thereby achieving personalized recommendations and improving the effectiveness and success rate of the collection.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a training method for a psychological portrait recognition model and a collection action recommendation system based on psychological portraits. Background Art

[0002] With the development of financial business, the number of customers who are overdue for repayment is also rising, which has brought great economic pressure to relevant financial institutions. The current collection methods mainly focus on superficial information such as the credit record and car purchase history of financial customers, and rarely consider the psychological characteristics of customers. This makes it impossible for collectors to accurately grasp the needs and reactions of customers, which can easily cause customers' resistance and reduce the collection effect; and the use of unified collection strategies and words lacks the ability of automatic learning and optimization, which affects the collection effect. If we can combine the psychological portrait of customers and provide targeted collection action recommendations, we can greatly improve the collection success rate.

[0003] Current psychological portrait recognition methods mostly rely on manual judgment by psychological experts, or on various psychological analysis theories combined with questionnaires. The classification is not accurate enough and does not have the ability to learn automatically. Summary of the invention

[0004] In view of the above problems, the present invention provides a training method for a psychological portrait recognition model and a collection action recommendation system based on psychological portraits, which overcome the above problems or at least partially solve the above problems. The technical solutions are as follows:

[0005] A training method for a psychological portrait recognition model, comprising:

[0006] Acquire a sample set, the sample set including score values ​​of N sample objects on J psychological evaluation labels respectively;

[0007] Forming a Euclidean space with a dimension of J using the J psychological evaluation labels, and determining the coordinates of each sample object in the sample set in the Euclidean space;

[0008] Determine K psychological profile categories, and call the allocation rule to allocate N sample objects in the sample set to a corresponding psychological profile category, wherein J, N and K are all positive integers greater than 1;

[0009] The allocation rules are:

[0010] Randomly select K centroids as the initial centers of K clusters, and update the initial centers of K clusters at least once;

[0011] Each time the update is performed, the coordinates of the centers of the current K clusters in the Euclidean space are determined, the Euclidean distances of each sample object in the sample set to the centers of the current K clusters are calculated, and each sample object in the sample set is assigned to the corresponding cluster with the shortest Euclidean distance; after all the assignments are completed, the weighted average of the coordinates of all sample objects assigned to each cluster in the Euclidean space is calculated as the coordinates of the centers of the K clusters in the Euclidean space after the update;

[0012] The K clusters formed after the last update correspond to the K psychological portrait categories, and the N sample objects in the sample set are allocated to the K clusters corresponding to the K psychological portrait categories in the last update.

[0013] In the above method, optionally, after all the allocations are completed, the formula for calculating the weighted average of the coordinates of all sample objects allocated to each cluster in the Euclidean space is:

[0014]

[0015] where c kj is the weighted average of the kth cluster on the jth dimension, x knkj is the sample n in the kth cluster k The value in the jth dimension, is the average value of all samples in the kth cluster on the jth dimension, α is the weight decay factor, s kj is the sample standard deviation of the kth cluster in the jth dimension, N k is the number of samples in the kth cluster, k∈[1,K],j∈[1,J], n k ∈[1,N k ].

[0016] The above method optionally determines whether a preset termination condition is met after each update, and if so, terminates the update, otherwise continues with the next update; the preset termination condition includes at least one of reaching the maximum number of iterations, the cluster center remains unchanged before and after the update, and the reliability of the distribution result reaches a preset threshold.

[0017] In the above method, when determining the value of K, multiple candidate values ​​of K are obtained and substituted into the following formula to calculate the corresponding clustering reliability r: K , select the calculated cluster reliability r K The candidate value corresponding to the highest value is taken as the final value of K:

[0018]

[0019]

[0020] where n∈[1,N], N1 to N K are the number of samples from the first cluster to the Kth cluster respectively; ω11 and ω12 are weight terms, and ω11+ω12=1; a(n) is the weighted average distance of all samples in the cluster where sample n is located, and b(n) is the weighted average distance of all samples in the cluster closest to sample n.

[0021] A collection action recommendation system based on psychological profiling, comprising:

[0022] A data collection module is used to obtain the historical collection behaviors of sample objects and the scores of J psychological evaluation tags, as well as the historical collection behaviors of collection objects;

[0023] A psychological evaluation label score determination module is used to train a psychological feature model based on the historical collection behavior of the sample object and the scores on J psychological evaluation labels, and input the historical collection behavior of the collection object into the trained psychological feature model to obtain the score values ​​of the collection object on J psychological evaluation labels;

[0024] A psychological portrait determination module is used to train a psychological portrait recognition model by executing the training method of the psychological portrait recognition model described above, and input the score values ​​of the collection object on the J psychological evaluation labels into the trained psychological portrait recognition model to obtain the psychological portrait category of the collection object;

[0025] The collection behavior recommendation module is used to output a recommended behavior list according to the psychological profile category of the collection object.

[0026] Compared with the prior art, the present invention has the following advantages: in the training method of the psychological portrait recognition model provided by the embodiment of the present invention, the scores of different evaluation dimensions are mapped to the Euclidean space in the form of coordinates, and different objects are assigned to appropriate psychological portrait categories in combination with cluster analysis. The psychological portrait recognition process can realize automatic learning and optimization, thereby improving the accuracy of classification; further, the clustering algorithm is improved in the embodiment, and the influence of abnormal samples on classification is reduced by introducing a sample weight factor. By using clustering reliability, the reliability of the overall clustering is improved, and the generalization ability is improved; the embodiment of the present invention also provides a method for recommending collection actions using the psychological portrait recognition model obtained by the above training, which provides collection action recommendations that match the psychological characteristics of the collection object by identifying the psychological portrait category of the collection object, thereby realizing personalized recommendations and improving the effectiveness and success rate of collection.

[0027] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0029] Figure 1 A method flow chart of a method for training a psychological portrait recognition model provided by an embodiment of the present invention;

[0030] Figure 2 A method flow chart of a method for recommending collection actions based on psychological profiling provided by an embodiment of the present invention;

[0031] Figure 3 A structural block diagram of a collection action recommendation system based on psychological profiling provided in an embodiment of the present invention;

[0032] Figure 4 A structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0034] refer to Figure 1 FIG. 4 shows a method flow chart of a method for training a psychological portrait recognition model provided by an embodiment of the present invention. Figure 1 The method execution process shown is a feasible implementation scheme of the training method of the psychological portrait recognition model provided by the embodiment of the present invention. The method can be applied to various system platforms, and its execution subject is a server set in the system platform. The training method of the psychological portrait recognition model specifically includes:

[0035] Acquire a sample set, the sample set including score values ​​of N sample objects on J psychological evaluation labels respectively;

[0036] In the method provided by the embodiment of the present invention, a sample set is used to train a psychological portrait recognition model, the score values ​​on J psychological evaluation labels are used as model input, and the recognized psychological portrait category is used as model output.

[0037] Optionally, the N sample objects may be N historical debt collection customers, and the scores of the N sample objects on the J psychological evaluation labels may be obtained by a psychological expert evaluating the N sample objects, such as by a telephone interview, a questionnaire, or the like. Further, based on the MBTI personality theory framework, the four personality dimensions used by the MBTI may be used as the J psychological evaluation labels for scoring, and the four personality dimensions are: extroversion-introversion, feeling-intuition, thinking-emotion, and judgment-perception. In the embodiment, the scores of the four personality dimensions of the N sample objects may be standardized by standard deviation to obtain the four personality dimension standard scores as the psychological characteristic labels of the N sample objects.

[0038] Forming a Euclidean space with a dimension of J using the J psychological evaluation labels, and determining the coordinates of each sample object in the sample set in the Euclidean space;

[0039] K psychological portrait categories are determined, and the allocation rule is called to allocate the N sample objects in the sample set to a corresponding psychological portrait category, wherein J, N and K are all positive integers greater than 1.

[0040] In the method provided by the embodiment of the present invention, the score value of each sample object on J psychological evaluation labels is mapped to a Euclidean space with a dimension of J. Combined with the characteristics of the Euclidean space, a cluster analysis is performed in the Euclidean space so that each sample object can be assigned to a psychological portrait category.

[0041] Furthermore, the allocation rule is: randomly select K centroids as the initial centers of K clusters, and update the initial centers of the K clusters at least once; determine the coordinates of the centers of the current K clusters in the Euclidean space in each update, calculate the Euclidean distance of each sample object in the sample set to the center of the current K clusters, and allocate each sample object in the sample set to the corresponding cluster with the closest Euclidean distance; after all allocations are completed, calculate the weighted average of the coordinates of all sample objects allocated to each cluster in the Euclidean space as the coordinates of the centers of the K clusters after the update in the Euclidean space; the K clusters formed after the last update correspond to the K psychological portrait categories, and the N sample objects in the sample set are allocated to the K clusters corresponding to the K psychological portrait categories in the last update.

[0042] In the method provided by the embodiment of the present invention, when calling the allocation rule to allocate the N sample objects in the sample set to the corresponding psychological portrait categories, K centroids are first randomly selected to obtain the initial centers of the K clusters, and the coordinates of the initial centers of the K clusters in the Euclidean space are determined. Then, for each sample object, the Euclidean distance between it and the initial centers of the K clusters is calculated, and the sample is allocated to the cluster with the closest distance, completing the first allocation. After the first allocation, the cluster center is updated for the first time. For each cluster, the weighted average of all samples in the cluster obtained through the first allocation is calculated, and the weighted average is used as the cluster center after the first update. After the first update, each sample object needs to be reallocated to the cluster with the closest Euclidean distance after the first update, completing the second allocation.

[0043] In the method provided by the embodiment of the present invention, at least one update is performed, and after each update, the N sample objects need to be reallocated to the K clusters finally updated. Further, optionally, after each update, it is determined whether a preset termination condition is met, and if so, the update is terminated, and the training of the psychological portrait recognition model is completed, and the K clusters formed after the last update correspond to the K psychological portrait categories, and the N sample objects in the sample set are allocated to the K clusters corresponding to the K psychological portrait categories in the last update. If the preset termination condition is not met, the next update is continued; the preset termination condition includes at least one of reaching the maximum number of iterations, the cluster center remains unchanged before and after the update, and the reliability of the distribution result is judged to reach a preset threshold.

[0044] If the next update is needed, such as the ith update, the weighted average of all samples in the cluster obtained after the ith allocation is calculated, and the weighted average is used as the cluster center after the ith update. Re-allocation is performed after each update, so the number of allocations is always one more than the number of updates. The psychological profile categories to which the N sample objects are finally allocated are determined based on the results of the last allocation.

[0045] In the method provided by the embodiment of the present invention, optionally, each time the cluster center is updated, the formula for calculating the weighted average value of the coordinates of the updated cluster center in the Euclidean space is:

[0046]

[0047] where c kj is the weighted average of the kth cluster on the jth dimension, x knkj is the sample n in the kth cluster k The value in the jth dimension, is the average value of all samples in the kth cluster on the jth dimension, s kj is the sample standard deviation of the kth cluster in the jth dimension, N kis the number of samples in the kth cluster, N1+N2+…+N K =N,k∈[1,K],j∈[1,J],n k ∈[1,N k ]. In the above formula, the second term on the right side of the equation is the correction term, is the sample weight term, α is the weight decay factor, and the larger α is, the faster the weight decays for samples that are farther from the original mean. The method provided in the embodiment of the present invention uses a weighted average to calculate the cluster center, which can effectively overcome the deviation of the cluster center caused by extreme abnormal samples.

[0048] In the method provided in the embodiment of the present invention, the value of K can be determined by calculating the clustering reliability. The specific method is: first, multiple candidate values ​​of K are obtained, and each candidate value of K is substituted into the following formula to calculate the corresponding clustering reliability r K , select the calculated cluster reliability r K The candidate value corresponding to the highest value is taken as the final value of K:

[0049]

[0050] where n∈[1,N], N1 to N K are the number of samples from the first cluster to the Kth cluster respectively; ω11 and ω12 are weight terms, and ω11+ω12=1; a(n) is the weighted average distance of all samples in the cluster where sample n is located, and b(n) is the weighted average distance of all samples in the cluster closest to sample n. Clustering reliability is composed of weighted silhouette coefficient and sample deviation. is the weighted contour coefficient term, sc(n) is the weighted contour coefficient of the nth sample, is the sample deviation. The larger the value, the more uneven the sample distribution. By trying different K candidate values, the final clustering reliability is maximized, thereby improving the recognition accuracy of the psychological portrait recognition model.

[0051] The present invention also provides a method for recommending collection actions by using the trained psychological portrait recognition model. Figure 2 FIG. 4 shows a flow chart of a method for recommending collection actions provided by an embodiment of the present invention. Figure 2 The method execution process shown is a feasible implementation scheme of the collection action recommendation method provided by an embodiment of the present invention. The collection action recommendation method provided by the present invention is based on psychological profiling and specifically includes:

[0052] Obtain a psychological portrait recognition model trained by the training method of the psychological portrait recognition model as described above; obtain the score value of the collection object on the J psychological evaluation labels, and determine the coordinates of the collection object in the Euclidean space; calculate the Euclidean distance between the collection object and the centers of the K clusters determined by the last update in the trained psychological portrait recognition model, and use the psychological portrait categories corresponding to the first M clusters with the closest distance as the psychological portrait categories of the collection object, where M is a positive integer and M∈(1,K); calculate the similarity sim(m) between the collection object and the M psychological portrait categories, obtain a set of recommended behaviors, calculate the weight value of each recommended behavior in the recommended behavior set in the M psychological portrait categories, arrange them from high to low according to the calculated weight values, and generate a corresponding recommended behavior recommendation list. When calculating the weight value of the recommended behavior in the M psychological portrait categories, the similarity sim(m) between the collection object and the M psychological portrait categories is introduced.

[0053] Furthermore, the calculation formula for the similarity sim(m) between the collection object and the mth psychological portrait category is:

[0054]

[0055] m∈[1,M], d(m) is the Euclidean distance between the collection object and the center of the cluster corresponding to the mth psychological profile category.

[0056] Furthermore, the calculation formula for the weight value of the recommended behavior in the M-type psychological portrait category is:

[0057] The weight value of the recommended behavior a in the mth psychological profile category

[0058] = the effectiveness of the recommended behavior a in the mth psychological profile category × sim(m)

[0059] In the method provided by the embodiment of the present invention, optionally, when calculating the weight value of each recommended behavior in the recommended behavior set in the M-type psychological portrait category, the historical usage of the recommended behavior set can be obtained, and the effectiveness of each recommended behavior in the corresponding psychological portrait category is calculated according to the historical usage of various recommended behaviors. The specific formula is:

[0060]

[0061] The weights of the recommended behaviors are calculated and sorted based on the effectiveness of each recommended behavior in the corresponding psychological portrait category and the similarity between the collection object and the psychological portrait category to obtain a sorted recommendation list.

[0062] In the method provided by the embodiment of the present invention, it is optional to obtain the score values ​​of the collection object on J psychological evaluation labels by establishing and training a psychological characteristic model. The specific method includes: obtaining the historical collection behavior records of N sample objects and the score values ​​of the N sample objects on J psychological evaluation labels as sample data of the psychological characteristic model, and training the psychological characteristic model; after obtaining the historical collection behavior of the collection object, inputting the trained psychological characteristic model, and outputting the score values ​​of the collection object on J psychological evaluation labels.

[0063] Optionally, the historical collection behavior records of N sample objects are obtained from the historical data, and the N sample objects can be scored on J psychological evaluation labels by psychological expert scoring, and the scores are standardized by standard deviation to obtain the standard scores of the N sample objects on J psychological evaluation labels. Furthermore, a text regression task is constructed based on the open source MacBERT pre-trained model (Revisiting Pre-trained Models for Chinese Natural Language Processing) to realize the construction of the psychological model. The implementation process is as follows: obtain the latest MacBERT pre-trained model parameters, load the model through the BertModel method in the transformers library, and define the model task as a regression task on J dimensions; split the sample data into a training set and a test set; define a loss function, which can be a mean square error loss function or a mean absolute error loss function; define an optimizer, which includes but is not limited to adamw, adam, sgd, adagrad, rmsprop, etc.; train the model to obtain a trained psychological feature model. The historical collection behavior is used as the input data of the psychological feature model, and the score value of the historical collection behavior object on J psychological evaluation labels is obtained after being processed by the psychological feature model. In the embodiment, a pre-trained model based on MacBert is used to construct a psychological model, which has a better effect than a traditional model based on a LIWC dictionary to predict psychological characteristics.

[0064] In the method provided by the embodiment of the present invention, the historical collection behavior of the collection object is obtained and input into the trained psychological feature model to obtain the score value of the collection object on J psychological evaluation labels; then the obtained score value of the collection object on the J psychological evaluation labels is mapped to the Euclidean space, and the Euclidean effect of the collection object and the Euclidean distance between the centers of the K clusters determined by the last update in the trained psychological portrait recognition model are calculated, the K clusters are sorted from near to far according to the Euclidean distance, and the psychological portrait categories corresponding to the first M clusters are selected as the psychological portrait categories of the collection object.

[0065] Optionally, to further improve the accuracy of recommendation, the psychological portrait categories corresponding to the first M clusters closest to each other can be used as the psychological portrait categories of the preliminarily determined collection targets, and then the M clusters can be further screened, and the psychological portrait categories corresponding to M1 clusters are selected as the final psychological portrait categories of the collection targets, where M1 is a positive integer not greater than M.

[0066] The method for selecting M1 clusters includes: respectively calculating the credibility of the psychological profile categories corresponding to the collection object and the M clusters, determining a threshold T, deleting the psychological profile categories corresponding to the M clusters whose credibility is not less than the threshold T, leaving M1 clusters, and using the psychological profile categories corresponding to the updated M1 clusters as the final psychological profile categories of the collection object; wherein the formula for calculating the credibility sa(m) of the psychological profile category corresponding to the collection object and the mth cluster is:

[0067]

[0068] ω21 and ω22 are weight terms, and ω21+ω22=1; s m is the standard deviation of the sample distance under the mth cluster. is the normalized distance term from the collection object to the mth cluster, is the normalized relative distance term from the collection object to the mth cluster, f m is the original relative distance, and

[0069] Calculate the similarity of the psychological portrait categories (M or M1) of the collection objects, and then calculate the weight value of the recommended behavior based on the similarity. Arrange the calculated weight values ​​from high to low to generate a corresponding recommended behavior recommendation list.

[0070] In the method provided in the embodiment of the present invention, the collection actions include telephone collection, online communication collection and face-to-face collection. The first two methods are usually used more frequently, and the recommended behaviors for collection actions are mainly dialogue script recommendations. Optionally, corresponding dialogue scripts or other collection actions can be refined from multiple levels (such as relationship formation, reality cognition, goal adjustment, resource resolution, commitment, etc.) based on psychological characteristics such as J psychological evaluation labels to form a default recommended behavior recommendation list. Business personnel can perform corresponding collection actions based on the recommended behavior recommendation list, or they can respond freely according to actual conditions, and record the collection actions implemented by business personnel for updating the recommended behavior library. The recommended behavior library can be automatically updated according to the actual collection results, greatly improving the collection efficiency and results.

[0071] In the method provided by the embodiment of the present invention, further, if the collection object has no historical collection behavior, it is impossible to obtain the score of the collection object on J psychological evaluation labels according to the psychological characteristic model, and it is also impossible to determine the psychological portrait category of the collection object; or when the psychological portrait category of the collection object cannot be determined according to the psychological portrait recognition model, a default recommended behavior recommendation list is output.

[0072] The present invention also provides a system for applying the above-mentioned collection action recommendation method based on psychological profiling, referring to Figure 3 The structure block diagram of a collection action recommendation system based on psychological portrait provided by an embodiment of the present invention is shown, and the collection action recommendation system specifically includes: a data acquisition module, a psychological evaluation label score determination module, a psychological portrait determination module and a collection behavior recommendation module, wherein the data acquisition module is used to obtain the historical collection behavior of the sample object and the score values ​​on J psychological evaluation labels, as well as the historical collection behavior of the collection object; the psychological evaluation label score determination module is used to train the psychological feature model according to the historical collection behavior of the sample object and the score values ​​on J psychological evaluation labels, and input the historical collection behavior of the collection object into the trained psychological feature model to obtain the score value of the collection object on J psychological evaluation labels; the psychological portrait determination module is used to execute the training method of the psychological portrait recognition model as described above to train the psychological portrait recognition model, and input the score value of the collection object on J psychological evaluation labels into the trained psychological portrait recognition model to obtain the psychological portrait category of the collection object; the collection behavior recommendation module is used to output a recommended behavior recommendation list according to the psychological portrait category of the collection object.

[0073] The system provided by an embodiment of the present invention may optionally further include a data center module and a collection behavior execution module, wherein the data center module communicates with the data acquisition module, the psychological evaluation label score determination module, the psychological portrait determination module and the collection behavior recommendation module, acquires and stores data provided by the data acquisition module, the psychological evaluation label score determination module, the psychological portrait determination module and the collection behavior recommendation module, and provides the data stored in the data center module to the data acquisition module, the psychological evaluation label score determination module, the psychological portrait determination module and the collection behavior recommendation module. The collection behavior execution module determines the collection object and provides the identity information of the collection object to the psychological evaluation label score determination module. After obtaining the identity information of the collection object, the psychological evaluation label score determination module obtains the historical collection behavior of the collection object from the data center module, inputs it into the trained psychological feature model to obtain the score value of the collection object on J psychological evaluation labels, and transmits it to the psychological portrait determination module; the psychological portrait determination module inputs the received score value of the collection object on J psychological evaluation labels into the trained psychological portrait recognition model, matches the psychological portrait category of the collection object to the collection behavior recommendation module, and the collection behavior recommendation module outputs the corresponding recommended behavior recommendation list to the collection behavior execution module. The collection behavior execution module provides the recommended behavior recommendation list to the business personnel, and the business personnel perform corresponding collection behaviors on the collection object according to the recommended behavior recommendation list, and the data acquisition module records the information of the collection behavior execution process and stores it in the data center module.

[0074] The collection action recommendation system provided in the embodiment of the present invention includes a processor and a memory. The above-mentioned collection action recommendation methods are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0075] The processor includes a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the password update process can be dynamically executed by adjusting the kernel parameters to improve the update efficiency.

[0076] An embodiment of the present invention provides a storage medium on which a program is stored. When the program is executed by a processor, the method for recommending collection actions is implemented.

[0077] An embodiment of the present invention provides a processor, which is used to run a program, wherein the collection action recommendation method is executed when the program is running.

[0078] like Figure 4As shown, an embodiment of the present invention provides an electronic device, the electronic device includes a bus, at least one processor connected to the bus, at least one memory, and at least one terminal; wherein the processor, the memory, and the terminal communicate with each other through the bus; the processor is used to call the program instructions in the memory to execute the above-mentioned collection action recommendation method, the terminal is used to collect collection actions and provide a recommended behavior recommendation list, and the terminal can optionally include a display, a microphone, a communication device, etc. Business personnel can use a set of terminal devices to perform collection actions and collect and store response data of collection objects.

[0079] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0080] In a typical configuration, the device includes one or more processors (CPU), memory and bus. The device may also include input / output interface, network interface and the like.

[0081] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0082] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0083] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0084] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0085] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A training method for a psychological portrait recognition model, characterized in that: include: Acquire a sample set, the sample set including score values ​​of N sample objects on J psychological evaluation labels respectively; Forming a Euclidean space with a dimension of J using the J psychological evaluation labels, and determining the coordinates of each sample object in the sample set in the Euclidean space; Determine K psychological profile categories, and call the allocation rule to allocate N sample objects in the sample set to a corresponding psychological profile category, wherein J, N and K are all positive integers greater than 1; The allocation rules are: Randomly select K centroids as the initial centers of K clusters, and update the initial centers of K clusters at least once; Each time the update is performed, the coordinates of the centers of the current K clusters in the Euclidean space are determined, the Euclidean distances of each sample object in the sample set to the centers of the current K clusters are calculated, and each sample object in the sample set is assigned to the corresponding cluster with the shortest Euclidean distance; after all the assignments are completed, the weighted average of the coordinates of all sample objects assigned to each cluster in the Euclidean space is calculated as the coordinates of the centers of the K clusters in the Euclidean space after the update; The K clusters formed after the last update correspond to the K psychological profile categories, and the N sample objects in the sample set are allocated to the K clusters corresponding to the K psychological profile categories in the last update; After all the assignments are completed, the formula for calculating the weighted average of the coordinates of all sample objects assigned to each cluster in the Euclidean space is: where c kj is the weighted average of the k-th cluster on the j-th dimension, is the sample n in the kth cluster k The value in the jth dimension, is the average value of all samples in the kth cluster on the jth dimension, α is the weight decay factor, s kj is the sample standard deviation of the kth cluster in the jth dimension, N k is the number of samples in the kth cluster, k∈[1,K],j∈[1,J], n k ∈[1,N k ].

2. The training method of the psychological portrait recognition model according to claim 1, characterized in that: After each update, it is determined whether the preset termination condition is met. If so, the update is terminated, otherwise the next update is continued; the preset termination condition includes at least one of reaching the maximum number of iterations, the cluster center remains unchanged before and after the update, and the reliability of the distribution result reaches a preset threshold.

3. The training method of the psychological portrait recognition model according to claim 1, characterized in that: When determining the value of K, obtain multiple candidate values ​​of K and substitute them into the following formula to calculate the corresponding clustering reliability r K , select the calculated cluster reliability r K The candidate value corresponding to the highest value is taken as the final value of K: where n∈[1,N], N1 to N K are the number of samples from the first cluster to the Kth cluster respectively; ω11 and ω12 are weight terms, and ω11+ω12=1; a(n) is the weighted average distance of all samples in the cluster where sample n is located, and b(n) is the weighted average distance of all samples in the cluster closest to sample n.

4. A collection action recommendation system based on psychological profiling, characterized in that: include: The data collection module is used to obtain the historical collection behaviors of the sample objects and the scores of the J psychological evaluation tags, as well as the historical collection behaviors of the collection objects; A psychological evaluation label score determination module is used to train a psychological feature model based on the historical collection behavior of the sample object and the scores on J psychological evaluation labels, and input the historical collection behavior of the collection object into the trained psychological feature model to obtain the score values ​​of the collection object on J psychological evaluation labels; A psychological portrait determination module, configured to execute the psychological portrait recognition model training method according to any one of claims 1 to 3 to train a psychological portrait recognition model, and input the score values ​​of the collection object on the J psychological evaluation labels into the trained psychological portrait recognition model to obtain the psychological portrait category of the collection object; The collection behavior recommendation module is used to output a recommended behavior list according to the psychological profile category of the collection object.

Images