Customer future demand prediction method and device, computer equipment and storage medium

Abstract

The invention relates to the field of model hosting, and provides a customer future demand prediction method and device, computer equipment and a computer readable storage medium, and the method comprises the steps: obtaining the current transaction data of a to-be-maintained customer, the current transaction data comprises personal basic information of the to-be-maintained customer and a current interaction record of a salesman and the to-be-maintained customer; encoding the current transaction data to obtain a feature vector; and inputting the feature vectors to the trained unsupervised learning model to obtain demand data of the to-be-maintained customer at a prediction time point, the unsupervised learning model being used for predicting feature changes and demand changes of the customer at different time points. By adopting the embodiment of the invention, the future demand prediction of the transacted users can be realized, the model is more accurate than the demand prediction model of the non-transacted customers, and the predicted demand data of the to-be-maintained customers at the prediction time point can provide high-quality services for the customers.

Description

Technical Field

[0001] The present application relates to the field of model hosting and financial technology, and in particular to a method, device, computer equipment and computer-readable storage medium for predicting future customer demand. Background Art

[0002] In today's insurance industry, personal insurance plays an important role. Its business model mainly relies on agents to conduct business activities. However, it faces many challenges and problems in this process. One of them is the prediction of customer demand.

[0003] Existing demand analysis tools are all based on the current status of customers or potential customers to predict demand and provide guidance and help to insurance sales personnel. However, since these customers have not yet been traded, the amount of information available is limited and difficult to convert into standardized information, which makes demand prediction difficult and less accurate. As for customers who have already been traded, practitioners currently lack the relevant maintenance awareness and have not developed corresponding demand analysis tools, resulting in a large number of customer losses.

[0004] Therefore, how to accurately predict the future needs of users who have already completed transactions and provide customers with high-quality services has become a technical problem that needs to be urgently solved in the insurance industry. Summary of the invention

[0005] The main purpose of this application is to provide a method, device, computer equipment and computer-readable storage medium for predicting customer future demand, aiming to realize the prediction of future demand of users who have completed transactions, so as to provide customers with high-quality services.

[0006] To achieve the above objectives, the present application provides a method for predicting future customer demand, which comprises the following steps:

[0007] Acquire current transaction data of the customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained;

[0008] Encoding the current transaction data to obtain a feature vector;

[0009] The feature vector is input into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0010] In addition, to achieve the above purpose, the present application also provides a customer future demand prediction device, the customer future demand prediction device comprising:

[0011] A data acquisition unit, used to acquire current transaction data of the customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained;

[0012] A data encoding unit, used for encoding the current transaction data to obtain a feature vector;

[0013] The demand prediction unit is used to input the feature vector into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0014] In addition, to achieve the above-mentioned purpose, the present application also provides a computer device, which includes a processor, a memory, and a customer future demand prediction program stored in the memory and executable by the processor, wherein when the customer future demand prediction program is executed by the processor, the steps of the customer future demand prediction method as described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a computer-readable storage medium, on which a customer future demand prediction program is stored, wherein when the customer future demand prediction program is executed by a processor, the steps of the customer future demand prediction method as described above are implemented.

[0016] The present application provides a method, apparatus, computer device and computer-readable storage medium for predicting future customer demand. The method comprises first obtaining current transaction data of a customer to be maintained, the current transaction data comprising basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained; then encoding the current transaction data to obtain a feature vector; and inputting the feature vector into a trained unsupervised learning model to obtain demand data of the customer to be maintained at a predicted time point. The unsupervised learning model is used to predict changes in characteristics and demand of the customer at different time points.

[0017] It can be seen that in the embodiment of the present application, the data used to predict the future needs of customers are the personal basic information of the customers to be maintained and the current interaction records between the salesperson and the customers to be maintained. Unlike the limited data that can be obtained for unsuccessful customers, the system stores a large amount of information about the customers who have been traded. By training the data of the high-value customers who have been traded, it can be well applied to predict the future needs of the customers who have been traded; secondly, the personal basic information of the customers and the interaction records between the salesperson and the customers are used in the unsupervised learning model, and the changes in the characteristics and needs of the customers at different time points can be learned through the unsupervised learning model, so as to establish a mapping relationship between the characteristics and the needs. Then, the demand data of the customers to be maintained at the future time point can be predicted based on the personal basic information of the customers to be maintained and the current interaction records between the salesperson and the customers to be maintained, and then the corresponding insurance services can be provided to the customers. It can be seen that the present application can realize the prediction of the future needs of the users who have been traded, and it is more accurate than the demand prediction model of the unsuccessful customers. The demand data of the customers to be maintained at the predicted time point predicted by the present application can provide customers with high-quality services. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for predicting future customer demand provided for this application;

[0019] Figure 2 A flow chart of another method for predicting future customer demand provided for this application;

[0020] Figure 3 A schematic diagram of the functional modules of a device for predicting future customer demand provided in this application;

[0021] Figure 4 A schematic block diagram of the structure of a computer device provided in this application.

[0022] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0025] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0026] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0027] The gateway platform-based traffic switching method involved in the embodiment of the present application is mainly applied to computer equipment. The gateway platform-based traffic switching generation device can be a PC, a portable computer, a mobile terminal, or other device with display and processing functions.

[0028] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0029] Reference Figure 1 , Figure 1 A flowchart of a method for predicting future customer demand provided for this application.

[0030] like Figure 1 As shown, an embodiment of the present application provides a method for predicting future customer demands, which includes steps S101 to S103.

[0031] In this embodiment, the method for predicting future customer demand includes the following steps:

[0032] Step S101: Acquire current transaction data of a customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained.

[0033] Among them, the customers to be maintained can be customers who have just completed a transaction, or customers whose activity value in the system is lower than or equal to the preset activity value. Exemplarily, the activity value can be determined based on the number of active days or active time points of the user. For example, if the number of active days is greater than or equal to one year, the active value is determined to be 100, if the number of active days is greater than or equal to half a year, the active value is determined to be 80, and if the number of active days is greater than or equal to one quarter, the active value is determined to be 60; for example, if the active time point is one week, the active value is determined to be 100, and if the active value is two weeks, the active value is determined to be 80, and so on. By predicting future demand for customers whose activity value is lower than or equal to the preset activity value, customers can be reminded at the corresponding time point, thereby activating customers and maintaining existing customers in the system.

[0034] Exemplarily, basic personal information may include name, gender, age, occupation, marital status, family status, education level, etc. The current interaction record may be, for example, how to contact the insurance company and the audio and video interaction record generated during the contact process.

[0035] Exemplarily, the current transaction data may be data recorded in the form of text, audio, video, picture, etc. If the current transaction data is audio data, the audio data may be converted into text data, and the speech emotion and intonation features in the audio data may be retained, and then encoded as part of the feature vector.

[0036] Exemplarily, when the current transaction data is image data, the features of the image data can be extracted through computer vision technology, such as using a deep convolutional neural network to recognize faces, expressions, actions, etc. in the image. The deep convolutional neural network can be, for example, ResNet or Inception.

[0037] For example, when the current transaction data is video data, video analysis techniques, such as motion recognition and scene understanding, can be used to analyze the non-verbal communication between the customer to be maintained and the salesperson, such as gestures, expressions, and body language. At the same time, the customer's image and emotional state features can be extracted in combination with the text data or voice data corresponding to the video data.

[0038] Step S102: Encode the current transaction data to obtain a feature vector.

[0039] Among them, the feature vector is a numerical representation that can be used to train models or perform data analysis. Encoding methods include, for example, at least one of the following: One-Hot Encoding, Label Encoding, and Word Embeddings. One-Hot Encoding converts each category value into a binary vector. For example, if there are three categories A, B, and C, they can be encoded as [1,0,0], [0,1,0], and [0,0,1]. Label encoding maps each category value to an integer. For example, A can be encoded as 0, B as 1, and C as 2. Word embedding uses pre-trained word vectors (such as Word2Vec, GloVe) to map words to vectors in a high-dimensional space.

[0040] Specifically, encoding the current transaction data to obtain a feature vector includes:

[0041] Performing text analysis on the current transaction data to obtain different types of current transaction data;

[0042] Performing semantic analysis on different types of current transaction data to obtain semantic features;

[0043] The semantic features are encoded to obtain a feature vector.

[0044] Among them, text analysis (Text Analysis), also known as text mining (Text Mining), refers to the process of extracting useful information and knowledge from a large amount of text data, aiming to understand and interpret the text content and discover the patterns and trends therein. The current transaction data includes the following: personal information, family status, product information, preference intentions. The classification method can be, for example, through clustering algorithms, long short-term memory networks (Long Short-Term Memory, LSTM), etc.

[0045] Among them, semantic analysis is an important task in natural language processing (NLP), which aims to understand the meaning and context of text. Semantic analysis not only involves the meaning at the lexical level, but also includes the meaning at the sentence and document levels. Through semantic analysis, key information, sentiment, topics, entities, etc. in the text can be extracted, thereby supporting various advanced applications such as sentiment analysis, question-answering systems, machine translation, text classification, etc.

[0046] Specifically, the text analysis of the current transaction data is performed to obtain different types of current transaction data, including:

[0047] Performing text analysis on the current transaction data for preprocessing to obtain serialized data, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and serialization;

[0048] The serialized data is input into a long short-term memory network to classify the current transaction data and obtain different types of the current transaction data.

[0049] Among them, text cleaning is an important step in text analysis and natural language processing, and its purpose is to remove noise and irrelevant information in the text so that subsequent processing and analysis are more efficient and accurate. For example, it can include removing punctuation marks, because punctuation marks usually do not have a direct impact on the understanding of the text content; it can also include removing stop words, which are common meaningless words such as "the", "is", "and", etc., and they can be removed using a predefined stop word list; it can also include removing blank characters, removing tabs and links, etc.

[0050] Among them, tokenization is the process of dividing text into words or phrases, which is a basic step in natural language processing. The result of tokenization is usually called "word unit" or "token". The quality of tokenization directly affects subsequent text processing tasks, such as word frequency statistics, word embedding, sentiment analysis, etc. Serialization is the process of converting the state information of an object into a form that can be stored or transmitted.

[0051] Among them, LSTM is a special recurrent neural network (RNN), which is specially designed to solve the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequence data. The LSTM network includes multiple key layers, each of which plays an important role in the entire network and works together to achieve accurate classification of the current transaction data.

[0052] Among them, the input layer is used to receive the serialized data. The length of the serialized data is the same, for example, it can include 8 dimensions or 16 dimensions, and the input layer passes the serialized information to the next LSTM layer. The LSTM layer is the core component of the neural network. Through its internal memory unit and gating mechanism, it effectively learns and extracts key information in the time series, thereby obtaining different demand changes at different times. At each time step, the LSTM layer uses the current input and previous memory state to update its internal state, and outputs different demand changes at different times. Then, the output is nonlinearly transformed and feature extracted through the fully connected layer. Finally, through the SoftMax layer, the output is mapped to the probability of each category, so as to obtain the current transaction data under categories such as personal information, family status, product information, and preference intentions.

[0053] Specifically, the semantic analysis of different types of current transaction data to obtain semantic features includes:

[0054] Extracting context representations of different types of current transaction data through a BERT model to obtain multiple tokens corresponding to each piece of current transaction data;

[0055] Inputting multiple word units corresponding to each piece of the current transaction data into a classifier to obtain a semantic role label of each word unit;

[0056] The semantic feature corresponding to each piece of the current transaction data is determined according to the semantic role label of each word.

[0057] Among them, BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model proposed by Google in 2018. It is based on the Transformer model, especially its encoder part. The Transformer model replaces the traditional recurrent neural network (RNN) and convolutional neural network (CNN) through the self-attention mechanism, and performs well in processing long-distance dependencies.

[0058] For example, one of the current transaction data is an intelligent voice conversation between a salesperson and a customer to be maintained: "Sure, I prefer conservative insurance." Then the semantic analysis can be implemented through the corresponding code as follows:

[0059] import torch

[0060] from transformers import BertTokenizer,BertForSequenceClassification

[0061] #Load the pre-trained BERT model and word segmenter

[0062] tokenizer=BertTokenizer.from_pretrained('bert-base-uncased')

[0063] model=BertForSequenceClassification.from_pretrained('bert-base-uncased')

[0064] # Input text

[0065] text="Sure, I prefer conservative insurance."

[0066] # Participle

[0067] inputs=tokenizer(text,return_tensors='pt')

[0068] #Get the model's predictions

[0069] outputs = model(**inputs)

[0070] logits = outputs.logits

[0071] #Get the predicted sentiment label

[0072] predicted_class_id=logits.argmax().item()

[0073] print(f'Predicted Sentiment:{predicted_class_id}')

[0074] Through the above analysis, words such as "sure" and "prefer to" are positive, so their semantic features can be judged to be positive. At the same time, the category of the current transaction data is preference intention. Therefore, based on the current transaction data, it can be determined that the preference intention is conservative. Therefore, at the same time point and under the same circumstances, conservative insurance purchase recommendations can be given.

[0075] Step S103: input the feature vector into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0076] Among them, unsupervised learning is a type of machine learning, whose goal is to discover potential structures and patterns from unlabeled data. Unlike supervised learning, unsupervised learning does not require pre-defined labels or target variables. Unsupervised learning models are widely used in tasks such as data exploration, clustering, dimensionality reduction, and anomaly detection.

[0077] Furthermore, before inputting the feature vector into the trained unsupervised learning model, the method further includes:

[0078] Using an autoencoder model and a generative adversarial network to perform feature extraction and data enhancement, and construct the unsupervised learning model;

[0079] The historical transaction data of the customers who have completed the transaction is obtained, and the unsupervised learning model is trained by using the historical transaction data to obtain the trained unsupervised learning model.

[0080] Among them, the autoencoder is a neural network model that is mainly used to learn the compressed representation of data (i.e. encoding) and reconstruct the original data from the compressed representation (i.e. decoding). Autoencoders are very useful in unsupervised learning and can be used for tasks such as dimensionality reduction, feature learning, data denoising, and generative models.

[0081] Among them, Generative Adversarial Networks (GAN) can generate synthetic data with a distribution similar to real data through adversarial training of two neural networks, the generator and the discriminator. GAN has achieved remarkable results in the fields of image generation, data enhancement, style transfer, etc.

[0082] In the embodiment of the present application, an unsupervised learning model is constructed using an autoencoder model and a generative adversarial network. First, the historical transaction data is compressed by the autoencoder to extract features, and then new data samples are generated by GAN. The generated data samples are used together with the original data to train the model to increase the diversity and quantity of the data. Thus, data enhancement can be achieved.

[0083] Furthermore, the method further comprises:

[0084] When the customer type of the closed customer does not include the customer type of the customer to be maintained, the model parameters of the generative adversarial network are adjusted to make the unsupervised learning model adapt to the change of the customer type.

[0085] For example, with the development of the economy and the progress of society, new things are constantly emerging, and the current unsupervised learning model may not be applicable to the new era. At this time, the model can be adjusted to adapt to the data distribution of the new customer group and reduce field differences. For example, the new customer group is more inclined to buy large savings products. The model can be adjusted through techniques such as adversarial training to better adapt to this purchasing behavior. In addition, some layers of the model can be unlocked and fine-tuned using the data of the new customer group to adapt to the new insurance product. If the new product is a life insurance for young professionals, the model can be fine-tuned to better understand the risk preferences of this group.

[0086] Furthermore, the method further comprises:

[0087] When detecting that a life event occurs to the customer to be retained, determining whether the life event affects the needs of the customer to be retained;

[0088] If the life event affects the needs of the customer to be retained, determining the potential impact of the life event on the needs;

[0089] Determining whether the feature vector includes a feature vector corresponding to the life event;

[0090] If the feature vector includes the feature vector corresponding to the life event, adjusting the weight of the feature vector corresponding to the life event according to the potential impact degree;

[0091] If the feature vector does not include the feature vector corresponding to the life event, then the feature vector corresponding to the life event is added.

[0092] Exemplarily, life events may include marriage, house purchase, childbirth, etc. At this time, the potential impact of the life events on demand can be determined. For example, if it belongs to the following categories of life events, it is judged that the demand of the customer to be maintained is affected. The categories of life events include family structure changes, market changes, policy adjustments, natural disasters, health crises, etc. At the same time, the potential impact is determined according to the degree of contact with the customer, and the weight of the feature vector is proportional to the potential impact. For example, if it is a market change or policy adjustment, there is a certain lag, and the impact on the customer itself is relatively limited, so its potential impact is lower; if it is a family structure change, a natural disaster, or a health crisis, it will immediately make the customer feel a sense of crisis, so it has a greater impact on the customer himself, so its potential impact is higher.

[0093] Exemplarily, the above-mentioned life events may or may not be included in the unsupervised learning model. For example, whether one is married or has children may already exist in the basic personal information. Therefore, it is only necessary to adjust the weight of the feature vector corresponding to the life event according to the potential impact. However, for unpredictable things such as natural disasters and policy changes, which are not included in the basic personal information, it is necessary to add the feature vector corresponding to the life event to improve the prediction accuracy of the model.

[0094] The above embodiment introduces event-driven, and the event-driven model combined with customer life events can dynamically adjust the insurance demand forecast, making the forecast more accurate and timely. This dynamic adjustment capability enables insurance companies to capture key moments in customers' lives in a timely manner, provide timely insurance services and products, enhance customer experience, and improve brand image.

[0095] The present application provides a method, apparatus, computer device and computer-readable storage medium for predicting future customer demand. The method comprises first obtaining current transaction data of a customer to be maintained, the current transaction data comprising basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained; then encoding the current transaction data to obtain a feature vector; and inputting the feature vector into a trained unsupervised learning model to obtain demand data of the customer to be maintained at a predicted time point. The unsupervised learning model is used to predict changes in characteristics and demand of the customer at different time points.

[0096] It can be seen that in the embodiment of the present application, the data used to predict the future needs of customers are the personal basic information of the customers to be maintained and the current interaction records between the salesperson and the customers to be maintained. Unlike the limited data that can be obtained for unsuccessful customers, the system stores a large amount of information about the customers who have been traded. By training the data of the high-value customers who have been traded, it can be well applied to predict the future needs of the customers who have been traded; secondly, the personal basic information of the customers and the interaction records between the salesperson and the customers are used in the unsupervised learning model, and the changes in the characteristics and needs of the customers at different time points can be learned through the unsupervised learning model, so as to establish a mapping relationship between the characteristics and the needs. Then, the demand data of the customers to be maintained at the future time point can be predicted based on the personal basic information of the customers to be maintained and the current interaction records between the salesperson and the customers to be maintained, and then the corresponding insurance services can be provided to the customers. It can be seen that the present application can realize the prediction of the future needs of the users who have been traded, and it is more accurate than the demand prediction model of the unsuccessful customers. The demand data of the customers to be maintained at the predicted time point predicted by the present application can provide customers with high-quality services.

[0097] Reference Figure 2 , Figure 2A flowchart of another method for predicting future customer demand provided for this application.

[0098] like Figure 2 As shown, an embodiment of the present application provides a method for predicting future customer demands, which includes steps S201 to S216.

[0099] In this embodiment, the method for predicting future customer demand includes the following steps:

[0100] Step S201: Use the autoencoder model and the generative adversarial network to perform feature extraction and data enhancement to construct the unsupervised learning model.

[0101] Step S202: Obtain historical transaction data of customers who have completed transactions, and train the unsupervised learning model using the historical transaction data to obtain the trained unsupervised learning model.

[0102] Step S203: When the customer type of the closed customer does not include the customer type of the customer to be maintained, adjust the model parameters of the generative adversarial network to make the unsupervised learning model adapt to the change of the customer type.

[0103] Step S204: Acquire current transaction data of the customer to be maintained, where the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained.

[0104] Step S205: Perform text analysis on the current transaction data for preprocessing to obtain serialized data, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and serialization.

[0105] Step S206: input the serialized data into a long short-term memory network to classify the current transaction data to obtain different types of the current transaction data.

[0106] Step S207: extract context representations of different types of current transaction data through the BERT model to obtain multiple word units corresponding to each piece of current transaction data.

[0107] Step S208: input multiple word units corresponding to each piece of the current transaction data into a classifier to obtain a semantic role label for each word unit.

[0108] Step S209: Determine the semantic feature corresponding to each piece of the current transaction data according to the semantic role label of each word.

[0109] Step S210: Encode the semantic feature to obtain a feature vector.

[0110] Step S211: when it is detected that a life event occurs in the customer to be maintained, determining whether the life event affects the needs of the customer to be maintained.

[0111] Step S212: If the life event affects the needs of the customer to be maintained, determine the potential impact of the life event on the needs.

[0112] Step S213: Determine whether the feature vector includes the feature vector corresponding to the life event.

[0113] Step S214: If the feature vector includes the feature vector corresponding to the life event, adjust the weight of the feature vector corresponding to the life event according to the potential impact degree.

[0114] Step S215: If the feature vector does not include the feature vector corresponding to the life event, add the feature vector corresponding to the life event.

[0115] Step S216: input the feature vector into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0116] The specific implementation methods of each step in the embodiments of the present application refer to the specific implementation methods of each step in the above embodiments, which will not be repeated here.

[0117] See also Figure 3 , Figure 3 A schematic diagram of the functional modules of a device for predicting future customer demand provided in this application.

[0118] like Figure 3 As shown, the customer future demand prediction device includes:

[0119] The data acquisition unit 301 is used to acquire the current transaction data of the customer to be maintained, wherein the current transaction data includes the basic personal information of the customer to be maintained and the current interaction record between the salesperson and the customer to be maintained;

[0120] A data encoding unit 302, used to encode the current transaction data to obtain a feature vector;

[0121] The demand prediction unit 303 is used to input the feature vector into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0122] Specifically, in encoding the current transaction data to obtain a feature vector, the data encoding unit 302 is specifically used to:

[0123] Performing text analysis on the current transaction data to obtain different types of current transaction data;

[0124] Performing semantic analysis on different types of current transaction data to obtain semantic features;

[0125] The semantic features are encoded to obtain a feature vector.

[0126] Specifically, in the aspect of performing text analysis on the current transaction data to obtain different types of current transaction data, the data encoding unit 302 is specifically used to:

[0127] Performing text analysis on the current transaction data for preprocessing to obtain serialized data, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and serialization;

[0128] The serialized data is input into a long short-term memory network to classify the current transaction data and obtain different types of the current transaction data.

[0129] Specifically, in the aspect of performing semantic analysis on the current transaction data of different types to obtain semantic features, the data encoding unit 302 is specifically used to:

[0130] Extracting context representations of different types of current transaction data through a BERT model to obtain multiple tokens corresponding to each piece of current transaction data;

[0131] Inputting multiple word units corresponding to each piece of the current transaction data into a classifier to obtain a semantic role label of each word unit;

[0132] The semantic feature corresponding to each piece of the current transaction data is determined according to the semantic role label of each word.

[0133] Furthermore, before inputting the feature vector into the trained unsupervised learning model, the customer future demand prediction device further includes:

[0134] A model building unit 304 is used to use an autoencoder model and a generative adversarial network to perform feature extraction and data enhancement to build the unsupervised learning model;

[0135] The model training unit 305 is used to obtain the historical transaction data of the customers who have completed the transaction, and to train the unsupervised learning model through the historical transaction data to obtain the trained unsupervised learning model.

[0136] Furthermore, the device for predicting future customer demand also includes:

[0137] The model adjustment unit 306 is used to adjust the model parameters of the generative adversarial network when the customer type of the completed customer does not include the customer type of the customer to be maintained, so that the unsupervised learning model can adapt to the change of the customer type.

[0138] Furthermore, the data encoding unit 302 is used to:

[0139] When detecting that a life event occurs to the customer to be retained, determining whether the life event affects the needs of the customer to be retained;

[0140] If the life event affects the needs of the customer to be retained, determining the potential impact of the life event on the needs;

[0141] Determining whether the feature vector includes a feature vector corresponding to the life event;

[0142] If the feature vector includes the feature vector corresponding to the life event, adjusting the weight of the feature vector corresponding to the life event according to the potential impact degree;

[0143] If the feature vector does not include the feature vector corresponding to the life event, then the feature vector corresponding to the life event is added.

[0144] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0145] The above-mentioned device can be implemented in the form of a computer program. Figure 4 Runs on the computer device shown.

[0146] See also Figure 4 , Figure 4 A schematic block diagram of the structure of a computer device provided in this application. The computer device may be a server.

[0147] See also Figure 4 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0148] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any customer future demand prediction method.

[0149] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0150] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any customer future demand prediction method.

[0151] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0152] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0153] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0154] Acquire current transaction data of the customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained;

[0155] Encoding the current transaction data to obtain a feature vector;

[0156] The feature vector is input into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

[0157] In one embodiment, in encoding the current transaction data to obtain a feature vector, the processor is used to run a computer program stored in a memory to implement the following steps:

[0158] Performing text analysis on the current transaction data to obtain different types of current transaction data;

[0159] Performing semantic analysis on different types of current transaction data to obtain semantic features;

[0160] The semantic features are encoded to obtain a feature vector.

[0161] In one embodiment, in the aspect of performing text analysis on the current transaction data to obtain different types of current transaction data, the processor is used to run a computer program stored in a memory to implement the following steps:

[0162] Performing text analysis on the current transaction data for preprocessing to obtain serialized data, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and serialization;

[0163] The serialized data is input into a long short-term memory network to classify the current transaction data and obtain different types of the current transaction data.

[0164] In one embodiment, in the aspect of performing semantic analysis on different types of current transaction data to obtain semantic features, the processor is used to run a computer program stored in a memory to implement the following steps:

[0165] Extracting context representations of different types of current transaction data through a BERT model to obtain multiple tokens corresponding to each piece of current transaction data;

[0166] Inputting multiple word units corresponding to each piece of the current transaction data into a classifier to obtain a semantic role label of each word unit;

[0167] The semantic feature corresponding to each piece of the current transaction data is determined according to the semantic role label of each word.

[0168] In one embodiment, before inputting the feature vector into the trained unsupervised learning model, the processor is used to run a computer program stored in a memory to implement the following steps:

[0169] Using an autoencoder model and a generative adversarial network to perform feature extraction and data enhancement, and construct the unsupervised learning model;

[0170] The historical transaction data of the customers who have completed the transaction is obtained, and the unsupervised learning model is trained by using the historical transaction data to obtain the trained unsupervised learning model.

[0171] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0172] When the customer type of the closed customer does not include the customer type of the customer to be maintained, the model parameters of the generative adversarial network are adjusted to make the unsupervised learning model adapt to the change of the customer type.

[0173] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0174] When detecting that a life event occurs to the customer to be retained, determining whether the life event affects the needs of the customer to be retained;

[0175] If the life event affects the needs of the customer to be retained, determining the potential impact of the life event on the needs;

[0176] Determining whether the feature vector includes a feature vector corresponding to the life event;

[0177] If the feature vector includes the feature vector corresponding to the life event, adjusting the weight of the feature vector corresponding to the life event according to the potential impact degree;

[0178] If the feature vector does not include the feature vector corresponding to the life event, then the feature vector corresponding to the life event is added.

[0179] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any of the customer future demand forecasting methods provided in the embodiments of the present application.

[0180] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0181] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting future customer demand, characterized in that: The method for predicting future customer demand comprises the following steps: Acquire current transaction data of the customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained; Encoding the current transaction data to obtain a feature vector; The feature vector is input into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

2. The method for predicting future customer demand according to claim 1, characterized in that: The encoding of the current transaction data to obtain a feature vector includes: Performing text analysis on the current transaction data to obtain different types of current transaction data; Performing semantic analysis on different types of current transaction data to obtain semantic features; The semantic features are encoded to obtain a feature vector.

3. The method for predicting future customer demand according to claim 2, characterized in that: The performing text analysis on the current transaction data to obtain different types of current transaction data includes: Performing text analysis on the current transaction data for preprocessing to obtain serialized data, wherein the preprocessing includes at least one of the following: text cleaning, word segmentation, and serialization; The serialized data is input into a long short-term memory network to classify the current transaction data and obtain different types of the current transaction data.

4. The method for predicting future customer demand according to claim 3, characterized in that: The semantic analysis of the current transaction data of different types to obtain semantic features includes: Extracting context representations of different types of current transaction data through a BERT model to obtain multiple tokens corresponding to each piece of current transaction data; Inputting multiple word units corresponding to each piece of the current transaction data into a classifier to obtain a semantic role label of each word unit; The semantic feature corresponding to each piece of the current transaction data is determined according to the semantic role label of each word.

5. The method for predicting future customer demand according to any one of claims 1 to 4, characterized in that: Before inputting the feature vector into the trained unsupervised learning model, the method further includes: Using an autoencoder model and a generative adversarial network to perform feature extraction and data enhancement, and construct the unsupervised learning model; The historical transaction data of the customers who have completed the transaction is obtained, and the unsupervised learning model is trained by using the historical transaction data to obtain the trained unsupervised learning model.

6. The method for predicting future customer demand according to claim 5, characterized in that: The method further comprises: When the customer type of the closed customer does not include the customer type of the customer to be maintained, the model parameters of the generative adversarial network are adjusted to make the unsupervised learning model adapt to the change of the customer type.

7. The method for predicting future customer demand according to any one of claims 1 to 4, characterized in that: The method further comprises: When detecting that a life event occurs to the customer to be retained, determining whether the life event affects the needs of the customer to be retained; If the life event affects the needs of the customer to be retained, determining the potential impact of the life event on the needs; Determining whether the feature vector includes a feature vector corresponding to the life event; If the feature vector includes the feature vector corresponding to the life event, adjusting the weight of the feature vector corresponding to the life event according to the potential impact degree; If the feature vector does not include the feature vector corresponding to the life event, then the feature vector corresponding to the life event is added.

8. A device for predicting future customer demand, characterized in that: The customer future demand prediction device comprises: A data acquisition unit, used to acquire current transaction data of the customer to be maintained, wherein the current transaction data includes basic personal information of the customer to be maintained and current interaction records between the salesperson and the customer to be maintained; A data encoding unit, used for encoding the current transaction data to obtain a feature vector; The demand prediction unit is used to input the feature vector into the trained unsupervised learning model to obtain the demand data of the customer to be maintained at the predicted time point. The unsupervised learning model is used to predict the feature changes and demand changes of the customer at different time points.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a customer future demand prediction program stored in the memory and executable by the processor, wherein when the customer future demand prediction program is executed by the processor, the steps of the customer future demand prediction method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a customer future demand prediction program, wherein when the customer future demand prediction program is executed by a processor, the steps of the customer future demand prediction method according to any one of claims 1 to 7 are implemented.

Images