B-end institution contact time recommendation method and system

Abstract

The invention discloses a method and a system for recommending contact time and contact information of a B-end mechanism. The method comprises a recommended contact date list acquisition step, a response rate prediction model training step and a recommendation step. According to the invention, for a B2B2C platform type enterprise, the optimal contact time and mode are recommended for a B-end mechanism. According to the method, the labor cost can be reduced, the dependence on annotation data of business experts is reduced, the manpower and resource investment is effectively controlled, and the operation efficiency is improved. Besides, the optimal contact time can be dynamically calculated according to the actual change of the B-end mechanism contact amount and the B-end client contact data, the method can adapt to the change of a service scene in time, and the timeliness and accuracy of recommendation are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of B-end institution customer management, and in particular, relates to a method and system for recommending B-end customer contact time. Background Art

[0002] B2B2C (Business-to-Business-to-Consumer) is a business model that combines the characteristics of B2B and B2C, also known as the "business-to-business-to-consumer" model. This model combines the cooperation between enterprises with the services between enterprises and consumers, forming a complex but efficient business ecosystem. The first "B" represents the supplier enterprise that provides products or services; the middle "B" is usually a platform enterprise, which plays the role of integrating resources and building channels; the last "C" is the consumer. In this model, the supplier enterprise sells products or services to consumers through the platform provided by the platform enterprise, and the platform enterprise plays a key role in connecting and serving both ends.

[0003] The medical beauty industry has developed rapidly in recent years, among which medical beauty consulting companies (platform enterprises) play an important connecting role in the entire industry chain. Medical beauty consulting companies are mainly responsible for collecting the medical beauty needs of customers (consumers) and recommending suitable medical beauty institutions (supplier companies) to customers based on these needs. They play a key role in communication and coordination in the process from consultation to completion of medical beauty projects.

[0004] Under this business model, consultants of medical beauty consulting companies undertake multi-faceted communication work. They not only need to communicate closely with customers, inquire about the face-to-face consultation, and understand the customer's experience and needs in the medical beauty process, but also contact the cooperating medical beauty institutions, follow up on the user's visit to the hospital, and ensure that the customer can successfully complete the medical beauty project. However, as the business scale expands, some problems gradually emerge.

[0005] In the management of B-end agency customers (supplier companies), the choice of contact time with B-end agencies usually relies on manual experience and simple statistical analysis, but this method lacks scientificity and accuracy, and it is difficult to fully capture the real behavior patterns of B-end agencies. With the development of big data and machine learning technology, some studies have begun to try to predict the optimal contact time of B-end agencies through supervised learning algorithms, in order to improve the response rate and satisfaction of B-end agencies. However, these methods face a key problem: a large amount of labeled data is required to train the model. In actual business, it is often impossible to obtain enough labeled data from B-end agencies, and the acquisition cost of labeled data is high and the cycle is long, which greatly limits the application effect of supervised learning algorithms. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for recommending contact time for B-end institutions, which can partially solve or alleviate the above-mentioned deficiencies in the prior art, and can efficiently and accurately recommend the time and method for contacting B-end institution customers for platform-type enterprises, and also obtain a higher response rate.

[0007] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions: A first aspect of the present invention is to provide a method for recommending contact time and contact information of a B-end institution, which is characterized by comprising: Steps to obtain the recommended contact date list: Perform preliminary clustering of the daily contact data of B-side organizations and assign the contact data to different clusters; Optimize the distance of the contact data in each cluster and construct a global similarity matrix; Constructing a Laplace matrix based on the global similarity matrix, performing secondary clustering after eigendecomposition of the Laplace matrix, and obtaining the cluster center point of each cluster after clustering; Build a time series prediction model to predict the feature value of each day in the future, calculate the distance between the feature value and the center point of each cluster to divide the feature value into clusters; filter out a list of recommended contact dates based on the feature value distribution of the cluster; Response rate prediction model training steps: Extract the factors that affect the response rate of B-side institutions; The response rate of each B-end institution is counted in natural days, and historical data with a response rate ranking higher than a ranking threshold is selected as a data set, and the data set uses the influencing factors as features to construct a sample; training a response rate prediction model using the data set; Recommended steps: For each B-side organization, based on its recommended contact date list, the response rate prediction model is used to predict the dates in the recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

[0008] As an improvement, the K-means clustering algorithm is used to preliminarily cluster the daily contact data of B-side institutions.

[0009] As an improvement, the method of constructing a global similarity matrix includes: Calculate the DTW distance of each data in the cluster to construct the intra-cluster similarity matrix; Calculate the DTW distance of the inter-cluster time series to construct the inter-cluster similarity matrix; The intra-cluster similarity matrix and the inter-cluster similarity matrix are integrated into a global similarity matrix.

[0010] As an improvement, the method of constructing the Laplacian matrix based on the global similarity matrix includes: The elements of each row of the global similarity matrix are summed up, and the result constitutes a degree matrix; Take the inverse of the square root of each diagonal element of the degree matrix D to obtain the transition matrix; The transition matrix is ​​normalized to obtain the Laplace matrix.

[0011] As an improvement, K-means clustering algorithm is used for secondary clustering.

[0012] As an improvement, clusters meeting the contact requirements are screened out according to pre-set screening conditions; The future dates belonging to the cluster are determined as recommended dates and a recommended date list is created.

[0013] As an improvement, the influencing factors include contact time period, contact method, number of contacts, average contact response time, average contact interval time, and customer stratification.

[0014] As an improvement, for a certain C-end customer, the response rate prediction model of the B-end institution is ; Among them, R ij is the response rate of the ith B-end institution on the jth day, C ij is the number of times the i-th B-end institution was contacted on the j-th day, in the morning ij ,afternoon ij ,night ij is the time period when the i-th B-end institution is contacted on the j-th day, I ij is the average contact time interval of the i-th B-end institution on the j-th day, ij Indicates that the contact method of the i-th B-end organization on the j-th day is system contact, group chat ij represents that the contact method of the i-th B-end organization on the j-th day was group chat contact, T ij is the average response time of the ith B-end institution on the jth day, and the general intention ij , low intention ij , high intention ij To segment the customers of this C-end customer, is the error term. ; a 1 、a 2 、a 3 、b 1 、b 2 、c 1 d 1 、e 1 、f 1 、f 2 、f3 All are weights.

[0015] The present invention also provides a B-end institution contact time and contact information recommendation system, comprising: The module for obtaining the recommended contact date list is used to obtain the recommended contact date list of the B-end institution; Response rate prediction model training module, used to train the response rate prediction model of B-side institutions; A recommendation module is used to obtain recommended contact time and contact method by using a recommended contact date list and a response rate prediction model.

[0016] As an improvement, the recommended contact date list acquisition module performs preliminary clustering on the daily contact data of the B-end organization and distributes the contact data to different clusters; Optimize the distance of the contact data in each cluster and construct a global similarity matrix; Constructing a Laplace matrix based on the global similarity matrix, performing secondary clustering after eigendecomposition of the Laplace matrix, and obtaining the cluster center point of each cluster after clustering; Build a time series prediction model to predict the feature value of each day in the future, calculate the distance between the feature value and the center point of each cluster to divide the feature value into clusters; filter out a list of recommended contact dates based on the feature value distribution of the cluster; The response rate prediction model training module extracts the influencing factors that affect the response rate of the B-end institution; The response rate of each B-end institution is counted in natural days, and historical data with a response rate ranking higher than a ranking threshold is selected as a data set, and the data set uses the influencing factors as features to construct a sample; training a response rate prediction model using the data set; For each B-end institution, the recommendation module uses a response rate prediction model to predict the dates in the recommended contact date list based on the recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

[0017] Beneficial effects: The present invention is aimed at B2B2C platform-type enterprises and recommends the best contact time and method for B-end institutions. First, obtain the recommended contact date list, use K-means to preliminarily cluster the contact data, build a global similarity matrix through DTW optimization distance, and then obtain the cluster center points through spectral clustering secondary clustering, and filter out the recommended dates in combination with time series prediction. Then train the response rate prediction model, extract the factors that affect the response rate of the institution, and construct a data set for training. For each B-end institution, according to the recommended date list, use the model to predict the time period and method with the highest response rate as a recommendation.

[0018] The present invention can reduce labor costs, reduce reliance on business experts’ labeled data, effectively control human and resource input, and improve operational efficiency. In addition, the present invention can dynamically calculate the optimal contact time based on the actual B-end institution contact volume and the changes in B-end customer contact data, timely adapt to changes in business scenarios, and ensure the timeliness and accuracy of recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.

[0020] Figure 1 It is a flow chart of the present invention.

[0021] Figure 2 This is a schematic diagram of the data after preliminary clustering.

[0022] Figure 3 Schematic diagram of cluster 1 after preliminary clustering.

[0023] Figure 4 Schematic diagram of the intra-cluster similarity matrix.

[0024] Figure 5 Schematic diagram of the global similarity matrix.

[0025] Figure 6 This is a schematic diagram of historical response rate data. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] Herein, suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings by themselves. Therefore, "module", "component" or "unit" can be used mixedly.

[0028] In this document, the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0029] In this document, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Herein "and / or" includes any and all combinations of one or more of the associated listed items.

[0031] Herein, "plurality" means two or more than two, ie, it includes two, three, four, five, etc.

[0032] Embodiment 1: like Figure 1 As shown, this embodiment provides a method for recommending contact time and contact information of a B-side institution, including: Steps to obtain the S1 recommended contact date list: S11 performs preliminary clustering of the daily contact data of B-side organizations and assigns the contact data to different clusters; Specifically, in this embodiment, the K-means clustering algorithm is used to perform preliminary clustering on the daily contact data of the B-side organization.

[0033] Use the K-means clustering algorithm to perform preliminary clustering on the daily contact data of each B-end organization (the elements of the sample include the number of contacts, average contact interval, average response time, change rate of the number of contacts, change rate of the average number of intervals, number of contacts, average number of intervals, and average response time), and quickly assign data points to different clusters, such as Figure 2 shown.

[0034] K-means is an unsupervised learning algorithm based on distance metric. Its basic idea is to divide data points into K different clusters in an iterative way so that the sum of the distances from each data point to the center point of its cluster is minimized. Through preliminary clustering, we can quickly understand the approximate distribution of the data and discover potential patterns and regularities in the data. For example, we may find that data points in certain clusters have a higher number of contacts and a shorter average response time, which may represent a class of active and efficient B-side organizations.

[0035] S12 optimizes the distance of the contact data in each cluster and constructs a global similarity matrix.

[0036] Since the Euclidean distance used by K-means is insensitive to the morphological changes of time series, the morphological characteristics of the time series of the data may be ignored. In the actual B-end institutional contact data, the morphology of the time series (such as the changing trend of the contact volume, etc.) may contain important information. Therefore, in order to more accurately measure the similarity between time series, this step introduces dynamic time warping technology (DTW) to optimize the distance calculation and construct a global similarity matrix that can more accurately reflect the data similarity relationship. DTW is a method for calculating the similarity between two time series. It uses the idea of ​​dynamic programming to find the optimal alignment between two time series, thereby overcoming the limitation that the Euclidean distance can only be compared point by point, and can better handle the expansion and deformation of the time series on the time axis. The specific steps include: S121 calculates the DTW distance of each data in the cluster to construct an intra-cluster similarity matrix.

[0037] like Figure 3 As shown, taking the dates 2024-10-10, 2024-10-14, and 2024-10-17 in cluster 1 as an example, the similarity matrix of these three dates is constructed. Since the length of each time series data in the cluster is 1, the multi-dimensional Euclidean distance between two points can be directly calculated. Example (in order to intuitively demonstrate the calculation process, the eigenvalues ​​are not normalized): .

[0038] Similarly, the distances between the three dates are calculated two by two, and finally the similarity matrix of the three dates is formed as follows: Figure 4 shown.

[0039] The intra-cluster similarity matrix records the similarity between each data point in the cluster. The smaller the element value in the matrix, the more similar the corresponding two data points are. Through the intra-cluster similarity matrix, the distribution and similarity of the data in the cluster can be intuitively seen, providing a basis for subsequent clustering refinement.

[0040] S122 calculates the DTW distance of the inter-cluster time series for constructing an inter-cluster similarity matrix.

[0041] Similarly, calculate the distance between different clusters, such as the DTW distance between 2024-10-10 in cluster 1 and 2024-10-12 in cluster 3, and form a similarity matrix between clusters.

[0042] The inter-cluster similarity matrix reflects the similarity of data points between different clusters. By analyzing the inter-cluster similarity matrix, we can understand the degree of association between different clusters, which helps to determine the boundaries and relationships between different clusters and provide important information for subsequent global clustering analysis.

[0043] S123 integrates the intra-cluster similarity matrix and the inter-cluster similarity matrix into a global similarity matrix.

[0044] The global similarity matrix covers the similarity information between all data points (including within and between clusters), providing a unified and comprehensive data basis for subsequent clustering refinement using methods such as spectral clustering. Based on this global matrix, the nonlinear structure and global relationship in the data can be captured more accurately, thereby obtaining more refined and reasonable clustering results. Figure 5 shown.

[0045] S13 constructs a Laplace matrix based on the global similarity matrix, performs secondary clustering after eigendecomposition of the Laplace matrix, and obtains the cluster center point of each cluster after clustering.

[0046] After preliminary clustering and distance optimization of the B-side institutional contact data, in order to divide the data more accurately and capture the nonlinear structure and global relationships in the data, the spectral clustering method is used to refine the clustering.

[0047] Spectral clustering is a clustering algorithm based on graph theory. It regards each data point in the data set as a node in the graph, and uses the similarity between data points as the weight of the edge between nodes, thereby constructing a weighted undirected graph. Clustering is achieved by performing eigendecomposition on the Laplacian matrix of the graph and analyzing its eigenvalues ​​and eigenvectors.

[0048] More specifically, the specific steps of constructing the Laplacian matrix based on the global similarity matrix include: S131 sums up each row element of the global similarity matrix, and the obtained result constitutes the degree matrix D.

[0049] The degree matrix D reflects how closely each node (data point) is connected to other nodes. The elements on the diagonal of the matrix represent the degree of the corresponding node, which is the sum of the similarities between the node and all other nodes.

[0050] S132 takes the inverse of the square root of each diagonal element of the degree matrix D to obtain a transition matrix.

[0051] Taking the inverse of the square root of each diagonal element of the degree matrix D, we get a new diagonal matrix, which can be called the transition matrix (usually denoted mathematically as ). Prepare for the subsequent calculation of the normalized Laplace matrix. By performing the above transformation on the degree matrix, the degree of the node can be standardized, making the weights of different nodes in subsequent calculations more comparable.

[0052] S133 performs normalization calculation on the transition matrix to obtain a Laplace matrix.

[0053] The normalized Laplace matrix L comprehensively considers the similarity between nodes and the degree of the nodes themselves, which can more effectively reflect the structural characteristics of the graph and provide an important basis for subsequent feature decomposition and clustering operations.

[0054] After obtaining the Laplacian matrix, the Laplacian matrix is ​​subjected to eigendecomposition.

[0055] Eigen decomposition can transform the information of the Laplacian matrix. The eigenvectors represent a certain potential structure of the data, and the eigenvalues ​​reflect the importance of these structures. By analyzing the eigenvalues ​​and eigenvectors, key patterns and relationships in the data can be found.

[0056] Secondary clustering (which can also be performed using the K-means clustering algorithm) can further refine the previous clustering results, making the data in each cluster more similar. The cluster center represents the typical characteristics of each cluster, and these cluster centers can be used to determine the cluster to which new data (such as feature values ​​in the future) belongs, thereby filtering out information such as recommended contact dates.

[0057] S14 builds a time series prediction model to predict the characteristic values ​​of each day in the future, calculates the distance between the characteristic value and the center point of each cluster to divide the characteristic value into clusters; and selects a list of recommended contact dates based on the characteristic value distribution of the cluster.

[0058] A time series forecasting model is a model used to analyze time series data and predict future values. In this scenario, a model is built based on historical B-end institution contact data (such as contact volume, average contact interval, and other features), and the model is used to predict these feature values ​​every day in the future. Common time series forecasting models include ARIMA (autoregressive integrated moving average model), LSTM (long short-term memory network), etc.

[0059] In the above clustering step, the center point of each cluster has been obtained. For the predicted feature value of each day in the future, calculate its distance from the center point of each cluster. According to the distance, the feature value is divided into the cluster with the closest distance, so as to determine its cluster affiliation.

[0060] Different clusters may represent different contact patterns. For example, some clusters may correspond to situations where the institution has a high response rate and a suitable amount of contact, while some clusters may correspond to situations where the institution is busy and has a low response rate. By analyzing the distribution of eigenvalues ​​within each cluster, we can select clusters that are more suitable for contact, select the future dates belonging to these clusters, and form a list of recommended contact dates.

[0061] More specifically, the specific steps of filtering out the recommended contact date list according to the characteristic value distribution of the cluster include: S141 selects clusters that meet the connection requirements according to the pre-set screening conditions.

[0062] Specifically, from the clusters obtained by clustering, the cluster with "contact volume_low, average contact interval_low, average response time_low, contact volume change rate_low, average interval number change rate_low, contact volume_rolling average 7 days_low, average interval number_rolling average 7 days_low, average response time_rolling average 7 days_low" is selected as the cluster that meets the requirements. In this embodiment, the selected cluster is a cluster that represents a small amount of contacts.

[0063] S142 determines the future dates belonging to the cluster as recommended dates and creates a recommended date list.

[0064] If a future day is predicted to belong to the above cluster with less representational connections through the time series prediction model, then this date will be used as the recommended date and aggregated into a recommended date list.

[0065] S2 response rate prediction model training steps: For the business scenarios of platform-based enterprises in B2B2C (customer service will contact the docking institutions and corresponding customers), different customers will be recommended to different institutions, and the customer service of platform-based enterprises will contact the institutions to obtain the customer's situation. For different customers, contact time and contact method, the response of the institution sometimes cannot meet our needs, so we need to calculate in what scenario to contact the institution, the response rate of the institution is high.

[0066] S21 extracts the factors that affect the response rate of B-side institutions.

[0067] like Figure 6As shown in the figure, through the analysis of the actual business situation, it is found that the response rate of the organization is related to multiple factors. The busyness and response willingness of the organization may be different in different contact time periods (morning, afternoon, evening); the contact method (such as system contact, group chat contact, etc.) will affect the effect of information transmission and the handling method of the organization; too many contacts may cause the organization to be bored, and too few contacts may not attract enough attention; the average contact response time reflects the organization's previous response efficiency; the average contact interval reflects the frequency of contact; customer stratification (such as general intention, low intention, high intention) will also affect the importance of the organization to the contact.

[0068] Therefore, in this embodiment, the above elements are used as factors affecting the response rate of the B-end mechanism.

[0069] S22 counts the response rate of each B-end institution on a natural day basis, selects historical data with a response rate ranking higher than a ranking threshold as a data set, and constructs a sample using the influencing factors as features.

[0070] The purpose of selecting historical data with response rate rankings above the ranking threshold (e.g., top 3) as the data set is to focus on situations where the organization has a good response, allowing the model to learn the patterns of factors that enable the organization to respond highly, thereby improving the accuracy of the model in predicting high response rate scenarios.

[0071] In this step, the influencing factors extracted in the previous step are used as features and combined with the corresponding response rate data to construct training samples. Each sample contains the specific value of the influencing factor and the corresponding institution response rate, providing data support for model training.

[0072] S23 uses the data set to train a response rate prediction model.

[0073] In this embodiment, for a certain C-end customer, the response rate prediction model of the B-end institution is: ; Among them, R ij is the response rate of the ith B-end institution on the jth day, C ij is the number of times the i-th B-end institution was contacted on the j-th day, in the morning ij ,afternoon ij ,night ij is the time period when the ith B-end institution is contacted on the jth day (usually represented by a binary variable, such as morning contact, morning ij is 1, otherwise 0), I ij is the average contact time interval of the i-th B-end institution on the j-th day, ij Indicates that the contact method of the i-th B-end organization on the j-th day is system contact, group chat ijrepresents that the contact method of the i-th B-end organization on the j-th day was group chat contact, T ij is the average response time of the ith B-end institution on the jth day, and the general intention ij , low intention ij , high intention ij To segment the customers of this C-end customer, It is the error term, which is used to represent the part that the model cannot fit accurately, reflecting the random factors and unconsidered factors in the actual situation.

[0074] a 1 、a 2 、a 3 、b 1 、b 2 、c 1 d 1 、e 1 、f 1 、f 2 、f 3 In the above steps, various factors affecting the response rate of B-side institutions were extracted and a data set was constructed to train a response rate prediction model. The purpose of using the semopy toolkit is to further construct and analyze the model, so as to clarify the specific influence of each feature (i.e. the number of contacts, contact time period, contact method and other influencing factors mentioned above) on the response rate of the institution, and provide a basis for the subsequent formulation of strategies to improve the response rate of the institution.

[0075] semopy is a Python toolkit for structural equation modeling (SEM). Structural equation modeling is a technique that uses a combination of statistical methods to analyze complex relationships between multiple variables. In this business scenario, it can help us estimate the weight parameters in the model and perform goodness-of-fit tests on the model to determine whether the model is reasonable and whether the relationship between each feature and the response rate is significant.

[0076] It should be noted that, in the present invention, step S1 and step S2 can be executed in any order and can be executed simultaneously.

[0077] S3 recommended steps: For each B-side organization, based on its recommended contact date list, the response rate prediction model is used to predict the dates in the recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

[0078] For each B-end institution, a list of recommended contact dates is obtained based on the previous steps. This list is screened after a series of operations such as preliminary clustering, distance optimization, secondary clustering, and time series prediction of the B-end institution contact data, and includes dates that are more suitable for contacting the institution in the future.

[0079] With the help of the previously trained response rate prediction model, we conduct an in-depth analysis of each date in the recommended contact date list. The model comprehensively considers the contact time period, contact method, number of contacts, average contact response time, average contact interval time, customer stratification and other factors that affect the response rate of B-end institutions. The model calculates the institution response rate corresponding to different time periods and different contact methods on each date.

[0080] The response rates corresponding to different time periods and contact methods on each date are compared, and the time period and contact method with the highest response rate are selected and determined as the recommended contact time and recommended contact method of the B-end agency respectively.

[0081] Finally, the system outputs recommendation results such as: The best contact time is the morning of 2024-10-12, using group chat.

[0082] Embodiment 2: This embodiment provides a B-end institution contact time and contact information recommendation system, including: The module for obtaining the recommended contact date list is used to obtain the recommended contact date list of the B-end institution; Response rate prediction model training module, used to train the response rate prediction model of B-side institutions; A recommendation module is used to obtain recommended contact time and contact method by using a recommended contact date list and a response rate prediction model.

[0083] More specifically, the recommended contact date list acquisition module performs preliminary clustering on the daily contact data of the B-end organization and distributes the contact data into different clusters; Optimize the distance of the contact data in each cluster and construct a global similarity matrix; Constructing a Laplace matrix based on the global similarity matrix, performing secondary clustering after eigendecomposition of the Laplace matrix, and obtaining the cluster center point of each cluster after clustering; Build a time series prediction model to predict the feature value of each day in the future, calculate the distance between the feature value and the center point of each cluster to divide the feature value into clusters; filter out a list of recommended contact dates based on the feature value distribution of the cluster; The response rate prediction model training module extracts the influencing factors that affect the response rate of the B-end institution; The response rate of each B-end institution is counted in natural days, and historical data with a response rate ranking higher than a ranking threshold is selected as a data set, and the data set uses the influencing factors as features to construct a sample; training a response rate prediction model using the data set; For each B-end institution, the recommendation module uses a response rate prediction model to predict the dates in the recommended contact date list based on the recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

[0084] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0085] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a computer terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0086] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for recommending contact time and contact information of a B-end institution, characterized in that include: Steps to obtain the recommended contact date list: Perform preliminary clustering of the daily contact data of B-side organizations and assign the contact data to different clusters; Optimize the distance of the contact data in each cluster and construct a global similarity matrix; Constructing a Laplace matrix based on the global similarity matrix, performing secondary clustering after eigendecomposition of the Laplace matrix, and obtaining the cluster center point of each cluster after clustering; Build a time series prediction model to predict the feature value of each day in the future, calculate the distance between the feature value and the center point of each cluster to divide the feature value into clusters; filter out a list of recommended contact dates based on the feature value distribution of the cluster; Response rate prediction model training steps: Extract the factors that affect the response rate of B-side institutions; The response rate of each B-end institution is counted in natural days, and historical data with a response rate ranking higher than a ranking threshold is selected as a data set, and the data set uses the influencing factors as features to construct a sample; training a response rate prediction model using the data set; Recommended steps: For each B-end organization, based on its recommended contact date list, the response rate prediction model is used to predict the dates in the recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

2. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that: The K-means clustering algorithm is used to preliminarily cluster the daily contact data of B-side institutions.

3. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that Methods for constructing a global similarity matrix include: Calculate the DTW distance of each data in the cluster to construct the intra-cluster similarity matrix; Calculate the DTW distance of the inter-cluster time series to construct the inter-cluster similarity matrix; The intra-cluster similarity matrix and the inter-cluster similarity matrix are integrated into a global similarity matrix.

4. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that Methods for constructing Laplacian matrices based on global similarity matrices include: The elements of each row of the global similarity matrix are summed up, and the result constitutes a degree matrix; Take the inverse of the square root of each diagonal element of the degree matrix D to obtain the transition matrix; The transition matrix is ​​normalized to obtain the Laplace matrix.

5. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that: The K-means clustering algorithm is used for secondary clustering.

6. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that Methods for filtering out a list of recommended contact dates based on the distribution of cluster eigenvalues ​​include: According to the pre-set screening conditions, clusters that meet the contact requirements are screened out; The future dates belonging to the cluster are determined as recommended dates and a recommended date list is created.

7. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that: The influencing factors include contact time period, contact method, number of contacts, average contact response time, average contact interval time, and customer stratification.

8. A method for recommending contact time and contact information of a B-end institution according to claim 1, characterized in that: For a certain C-end customer, the response rate prediction model of the B-end organization is: ； Among them, R ij is the response rate of the ith B-end institution on the jth day, C ij is the number of times the i-th B-end institution was contacted on the j-th day, in the morning ij ,afternoon ij ,night ij is the time period when the i-th B-end institution is contacted on the j-th day, I ij is the average contact time interval of the i-th B-end institution on the j-th day, ij Indicates that the contact method of the i-th B-end organization on the j-th day is system contact, group chat ij represents that the contact method of the i-th B-end organization on the j-th day was group chat contact, T ij is the average response time of the ith B-end institution on the jth day, and the general intention ij , low intention ij , high intention ij To segment the customers of this C-end customer, is the error term; a1, a2, a3, b1, b2, c1, d1, e1, f1, f2, f3 are all weights.

9. A B-end institution contact time and contact information recommendation system, characterized in that include: The module for obtaining the recommended contact date list is used to obtain the recommended contact date list of the B-end institution; Response rate prediction model training module, used to train the response rate prediction model of B-side institutions; A recommendation module is used to obtain recommended contact time and contact method by using a recommended contact date list and a response rate prediction model.

10. A B-end institution contact time and contact information recommendation system according to claim 9, characterized in that: The recommended contact date list acquisition module performs preliminary clustering on the daily contact data of the B-end organization and distributes the contact data to different clusters; Optimize the distance of the contact data in each cluster and construct a global similarity matrix; Constructing a Laplace matrix based on the global similarity matrix, performing secondary clustering after eigendecomposition of the Laplace matrix, and obtaining the cluster center point of each cluster after clustering; Construct a time series prediction model to predict the feature value of each day in the future, calculate the distance between the feature value and the center point of each cluster to divide the feature value into clusters; Filter out a list of recommended contact dates based on the distribution of cluster eigenvalues; The response rate prediction model training module extracts the influencing factors that affect the response rate of the B-end institution; The response rate of each B-end institution is counted in natural days, and historical data with a response rate ranking higher than a ranking threshold is selected as a data set, and the data set uses the influencing factors as features to construct a sample; training a response rate prediction model using the data set; For each B-end institution, the recommendation module uses a response rate prediction model to predict the dates in the recommended contact date list based on its recommended contact date list to obtain the time period and contact method with the highest response rate as the recommended contact time and contact method.

Images