Method for predicting customer purchase intention, related device and computer storage medium

Abstract

The application provides a method for predicting a purchase intention of a customer, a related device and a computer storage medium. The method comprises the following steps: firstly, reading all communication data of a customer today from a communication database; wherein the communication data comprises each conversation text of communication with the customer; then, for each customer, splicing each conversation text of communication with the customer to obtain today's communication text of the customer; and inputting the today's communication text of the customer into a prediction model to process the today's communication text of the customer by the prediction model to obtain a prediction result of a purchase intention of the customer today; wherein the prediction model is obtained by training a neural network model by a training sample set; the training sample set comprises at least one training sample text; and the training sample text comprises positive sample text and negative sample text. Thus, the purchase intention of the customer can be accurately predicted.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus and computer storage medium for predicting customer purchase intentions. Background Technology

[0002] Currently, when sales consultants assess a customer's purchasing intention, subjective experience is often the dominant factor. Therefore, situations arise where sales consultants lack experience or are unable to accurately judge the strength of a customer's buying intention, leading to customer loss and the loss of potential customers with high purchasing intent.

[0003] Current technologies also employ methods that predict the semantic intent of text to determine a customer's purchasing intention. However, for long conversational texts, a common approach is to use a sliding window, segmenting the long text into many pieces and feeding each piece into an encoder (such as BERT) to generate text vectors. This approach loses the connection between the segments. Furthermore, the sales process between salespeople and customers typically involves more than one conversational exchange. Current practices usually consider multiple conversational texts during model training but predict results based on the granularity of a single conversational text during prediction, failing to account for the textual connections between consecutive conversations. This results in inconsistencies between the input during model training and prediction, leading to inaccurate predictions. Summary of the Invention

[0004] In view of this, this application provides a method, related apparatus, and computer storage medium for predicting customer purchase intentions, for accurately predicting customer purchase intentions.

[0005] The first aspect of this application provides a method for predicting customer purchase intentions, including:

[0006] Read all customer communication data for today from the communication database; wherein, the communication data includes the text of each conversation with the customer;

[0007] For each customer, the text of each conversation with the customer is concatenated to obtain the customer's communication text for today;

[0008] The customer's communication text for today is input into the prediction model. The prediction model processes the customer's communication text for today to obtain a prediction result of the customer's purchase intention for today. The prediction model is obtained by training a neural network model using a training sample set. The training sample set includes at least one training sample text. The training sample text includes positive sample text and negative sample text. The positive sample text is defined as follows: if a customer makes a purchase within N days after a sample conversation, the sample conversation text and the first M sample conversation texts of the sample conversation text are considered as positive sample text. The negative sample text is defined as follows: if a customer makes no purchase within N days after a sample conversation, the sample conversation text and the first M sample conversation texts of the sample conversation text are considered as negative sample text.

[0009] Optionally, the step of processing the customer's communication text for today using the prediction model to obtain a prediction result of the customer's purchase intention for today includes:

[0010] The customer's communication text for today is preprocessed to obtain the preprocessed communication text for today.

[0011] The preprocessed customer communication text for today is filtered to obtain the target communication text;

[0012] Feature extraction is performed on the target communication text to obtain the feature vector of the target communication text;

[0013] The feature vector of the target communication text is fused with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text; wherein, the feature vector of the customer's historical target communication text is obtained by querying a historical feature vector database;

[0014] The feature vector of the fused target communication text is input into the prediction function to obtain the prediction result of the customer's purchase intention today.

[0015] Optionally, after fusing the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text, the method further includes:

[0016] The feature vector of the fused target communication text is used to replace the feature vector of the customer's historical target communication text in the historical feature vector database.

[0017] Optionally, the method for constructing the prediction model includes:

[0018] Construct a training sample set; wherein the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; the positive sample text is defined as: if a sample customer makes a purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as positive sample text; the negative sample text is defined as: if a sample customer does not make a purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as negative sample text.

[0019] In the training sample set, a training sample text is randomly selected;

[0020] The training sample text is input into a neural network model, which processes the training sample text and outputs a prediction result; wherein, the prediction result is whether the sample customer will make a purchase in the next N days;

[0021] The parameters in the neural network model are continuously adjusted using the error between the predicted result and the actual result until the error between the predicted result and the actual result output by the adjusted neural network model meets a preset convergence condition. The adjusted neural network model is then determined as the prediction model. Specifically, if the training sample is positive sample text, the actual result is that the sample customer will make a purchase within the next N days; if the training sample is negative sample text, the actual result is that the sample customer will not make a purchase within the next N days.

[0022] Optionally, processing the training sample text using the neural network model includes:

[0023] The training sample text is preprocessed to obtain the preprocessed training sample text.

[0024] The preprocessed training sample text is filtered to obtain the target training sample text.

[0025] Feature extraction is performed on the target training sample text to obtain the feature vector of the target training sample text;

[0026] The feature vector of the target training sample text is input into the prediction function to obtain the prediction result.

[0027] Optionally, after obtaining the predicted result of the customer's purchase intention for today, the output further includes:

[0028] Record the predicted purchase intentions of the customers today in the intention table.

[0029] A second aspect of this application provides a device for predicting customer purchase intentions, comprising:

[0030] The reading unit is used to read all customer communication data for today from the communication database; wherein, the communication data includes the text of each conversation with the customer;

[0031] The splicing unit is used to splice together the text of each conversation with the customer for each customer to obtain the customer's communication text for today.

[0032] The first prediction unit is used to input the customer's communication text for today into the prediction model, and process the customer's communication text for today through the prediction model to obtain the prediction result of the customer's purchase intention for today; wherein, the prediction model is obtained by training a neural network model with a training sample set; the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; the positive sample text is defined as: if there is a purchase behavior within N days after the sample dialogue text with the sample customer, then the sample dialogue text and the sample dialogue text of the first M times of the sample dialogue text are regarded as positive sample text; the negative sample text is defined as: if there is no purchase behavior within N days after the sample dialogue text with the sample customer, then the sample dialogue text and the sample dialogue text of the first M times of the sample dialogue text are regarded as negative sample text.

[0033] Optionally, when the first prediction unit processes the customer's communication text for today using the prediction model to obtain a prediction result of the customer's purchase intention for today, it includes:

[0034] The first preprocessing unit is used to preprocess the customer's communication text for today to obtain the preprocessed communication text for today.

[0035] The first filtering unit is used to filter the preprocessed customer's communication text of the day to obtain the target communication text;

[0036] The first feature extraction unit is used to extract features from the target communication text to obtain the feature vector of the target communication text;

[0037] The fusion unit is used to fuse the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text; wherein, the feature vector of the customer's historical target communication text is obtained by querying a historical feature vector database;

[0038] The first input unit is used to input the feature vector of the fused target communication text into the prediction function to obtain the prediction result of the customer's purchase intention today.

[0039] Optionally, the customer purchase intention prediction device further includes:

[0040] The replacement unit is used to replace the feature vector of the customer's historical target communication text in the historical feature vector database with the feature vector of the fused target communication text.

[0041] Optionally, the building blocks of the prediction model include:

[0042] A training sample set construction unit is used to construct a training sample set; wherein, the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; the positive sample text is defined as: if a sample customer makes a purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as positive sample text; the negative sample text is defined as: if a sample customer makes no purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as negative sample text.

[0043] A selection unit is used to randomly select a training sample text from the training sample set;

[0044] The second prediction unit is used to input the training sample text into a neural network model, process the training sample text through the neural network model, and output a prediction result; wherein, the prediction result is whether the sample customer will make a purchase in the next N days;

[0045] The model determination unit is used to continuously adjust the parameters in the neural network model using the error between the predicted result and the actual result, until the error between the predicted result and the actual result output by the adjusted neural network model meets a preset convergence condition, and then determines the adjusted neural network model as the prediction model; wherein, if the training sample is positive sample text, the actual result is that the sample customer will make a purchase in the next N days; if the training sample is negative sample text, the actual result is that the sample customer will not make a purchase in the next N days.

[0046] Optionally, when the second prediction unit processes the training sample text using the neural network model and outputs a prediction result, it includes:

[0047] The second preprocessing unit is used to preprocess the training sample text to obtain preprocessed training sample text.

[0048] The second filtering unit is used to filter the preprocessed training sample text to obtain the target training sample text.

[0049] The second feature extraction unit is used to extract features from the target training sample text to obtain the feature vector of the target training sample text.

[0050] The second input unit is used to input the feature vector of the target training sample text into the prediction function to obtain the prediction result.

[0051] Optionally, the customer purchase intention prediction device further includes:

[0052] The recording unit is used to record the predicted results of the customer's purchase intention for today into the intention table.

[0053] A third aspect of this application provides an electronic device, comprising:

[0054] One or more processors;

[0055] A storage device on which one or more programs are stored;

[0056] When the one or more programs are executed by the one or more processors, the one or more processors implement the customer purchase intention prediction method as described in any one of the first aspects.

[0057] The fourth aspect of this application provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for predicting customer purchase intentions as described in any one of the first aspects.

[0058] As can be seen from the above solutions, this application provides a method, related apparatus, and computer storage medium for predicting customer purchase intention. The method for predicting customer purchase intention includes: first, reading all customer communication data for today from a communication database; wherein, the communication data includes the text of each conversation with the customer; then, for each customer, concatenating the text of each conversation with the customer to obtain the customer's communication text for today; and then inputting the customer's communication text for today into a prediction model, processing the customer's communication text for today through the prediction model to obtain a prediction result of the customer's purchase intention for today. The prediction model is obtained by training a neural network model using a training sample set. The training sample set includes at least one training sample text, which comprises positive and negative sample texts. A positive sample text is defined as follows: if a customer makes a purchase within N days after a sample dialogue, the sample dialogue text and the first M samples of that dialogue are considered positive sample texts. A negative sample text is defined as follows: if a customer does not make a purchase within N days after a sample dialogue, the sample dialogue text and the first M samples of that dialogue are considered negative sample texts. This achieves the goal of accurately predicting customer purchase intentions. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0060] Figure 1 A flowchart illustrating a method for predicting customer purchase intentions provided in this application embodiment;

[0061] Figure 2 A flowchart illustrating a method for predicting customer purchase intentions, provided as another embodiment of this application;

[0062] Figure 3 This is a schematic diagram illustrating the preprocessing of training sample text according to another embodiment of this application;

[0063] Figure 4 A detailed flowchart illustrating a method for constructing a prediction device according to another embodiment of this application;

[0064] Figure 5A flowchart illustrating the processing of training sample text using a neural network model, as provided in another embodiment of this application;

[0065] Figure 6 A schematic diagram of a customer purchase intention prediction device provided for another embodiment of this application;

[0066] Figure 7 A schematic diagram of a first prediction unit provided for another embodiment of this application;

[0067] Figure 8 A schematic diagram of a construction unit for a prediction device provided in another embodiment of this application;

[0068] Figure 9 A schematic diagram of a second prediction unit provided for another embodiment of this application;

[0069] Figure 10 A schematic diagram of an electronic device for realizing a method for predicting customer purchase intentions, provided as another embodiment of this application. Detailed Implementation

[0070] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0071] It should be noted that the concepts of "first," "second," etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0072] This application provides a method for predicting customer purchase intention, such as... Figure 1 As shown, the specific steps include:

[0073] S101. Retrieve all customer communication data for today from the communication database.

[0074] The communication data includes the text of every conversation with the customer.

[0075] It should be noted that the communication database can store voice data from communications with customers, text of every conversation with customers, or voice data from communications with customers converted into text for storage; there are no restrictions here.

[0076] S102. For each customer, the text of each conversation with the customer is spliced ​​together to obtain the customer's communication text for today.

[0077] S103. Input the customer's communication text for today into the prediction model, process the customer's communication text for today through the prediction model, and obtain the prediction result of the customer's purchase intention for today.

[0078] The prediction model is obtained by training the neural network model using a training sample set. The training sample set includes at least one training sample text. The training sample text includes positive sample text and negative sample text. A positive sample text is defined as follows: if a customer makes a purchase within N days after engaging in a sample dialogue, the sample dialogue text and the first M samples of the sample dialogue text are considered as positive sample text. A negative sample text is defined as follows: if a customer makes a purchase within N days after engaging in a sample dialogue, the sample dialogue text and the first M samples of the sample dialogue text are considered as negative sample text.

[0079] Optionally, in another embodiment of this application, one implementation method involves processing the customer's communication text for today using a predictive model to obtain a predicted result of the customer's purchase intention for today, such as... Figure 2 As shown, it includes:

[0080] S201. Preprocess the customer's communication text for today to obtain the preprocessed communication text for today.

[0081] Specifically, the contextual information from the customer's communication text today is combined to form the segment containing the customer's original communication content; this is the pre-processed version of the customer's communication text today. For example... Figure 3 The images shown are fragment 1 and fragment 2. Each fragment contains dialogue between two roles: a customer and a sales consultant.

[0082] As can be seen, unlike existing technologies, this application does not directly extract long texts, but rather breaks them down into multiple segments, thereby ensuring the semantic coherence of the communication text between the customer and the sales / consultant.

[0083] S202. Filter the pre-processed customer communication text to obtain the target communication text.

[0084] Specifically, each pre-processed customer's daily communication text is processed by a pre-trained semantic point classification model. The main function of this model is to determine whether each pre-processed customer's daily communication text contains valid semantics and to filter out invalid semantic segments. This acts as a filter, preventing too many invalid semantic segments from affecting the overall encoding effect and achieving the goal of reducing segments and improving performance.

[0085] S203. Extract features from the target communication text to obtain the feature vector of the target communication text.

[0086] Specifically, feature extraction methods can be used, but are not limited to, BERT text feature encoding; no specific limitations are made here.

[0087] S204. The feature vector of the target communication text is fused with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text.

[0088] The feature vectors of the customer's historical target communication texts are obtained by querying the historical feature vector database.

[0089] It should be noted that the method of fusing the feature vector of the target communication text with the feature vector of the customer's historical target communication text can be, but is not limited to, the vector averaging method; no limitation is made here.

[0090] Optionally, in another embodiment of this application, an implementation method that fuses the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text further includes:

[0091] Replace the feature vectors of the customer's historical target communication text in the historical feature vector database with the feature vectors of the merged target communication text.

[0092] This application retains the customer's historical communication content and stores it as text feature vectors, avoiding the need to repeatedly encode historical communication content while maintaining consistency with the model training. This avoids the inconsistency caused by multiple continuous communication dialogues as input during training but only the communication dialogues of the current day during prediction.

[0093] S205. Input the feature vector of the fused target communication text into the prediction function to obtain the prediction result of the customer's purchase intention today.

[0094] Optionally, in another embodiment of this application, one implementation of the method for constructing the prediction model is as follows: Figure 4 As shown, it includes:

[0095] S401. Construct a training sample set.

[0096] The training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; a positive sample text is defined as: if a customer makes a purchase within N days after engaging in a sample dialogue with the customer, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are used as positive sample text; a negative sample text is defined as: if a customer makes a purchase within N days after engaging in a sample dialogue with the customer, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are used as negative sample text.

[0097] S402. Randomly select a training sample text from the training sample set.

[0098] It should be noted that after training the neural network model using randomly selected training sample texts, another randomly selected training sample text can be used to continue training the neural network model. It is understandable that the more training sample texts used, the more accurate the trained neural network model, i.e., the prediction model, will be.

[0099] S403. Input the training sample text into the neural network model, process the training sample text through the neural network model, and output the prediction result.

[0100] The prediction result is whether the sample customers will make a purchase within the next N days.

[0101] Optionally, in another embodiment of this application, one approach is to process the training sample text using a neural network model, such as... Figure 5 As shown, it includes:

[0102] S501. Preprocess the training sample text to obtain the preprocessed training sample text.

[0103] Specifically, the contextual communication content between the training sample text and the sample customer's communication content is combined to form the segment containing the current sample customer's communication content; that is, the preprocessed training sample text. Similarly, it can be done as follows: Figure 3 The images shown are fragment 1 and fragment 2. Each fragment contains dialogue between two roles: a customer and a sales consultant.

[0104] As can be seen, unlike existing technologies, this application does not directly extract long texts, but rather breaks them down into multiple segments, thereby ensuring the semantic coherence of the communication text between the customer and the sales / consultant.

[0105] S502. Filter the preprocessed training sample text to obtain the target training sample text.

[0106] Specifically, each preprocessed training sample text is passed through a pre-trained semantic point classification model. The main function of this model is to determine whether each preprocessed training sample text contains valid semantics and to filter out invalid semantic segments. This acts as a filter, preventing too many invalid semantic segments from affecting the overall encoding effect and achieving the goal of reducing segments and improving performance.

[0107] S503. Extract features from the target training sample text to obtain the feature vector of the target training sample text.

[0108] Specifically, feature extraction methods can be used, but are not limited to, BERT text feature encoding; no specific limitations are made here.

[0109] S504. Input the feature vector of the target training sample text into the prediction function to obtain the prediction result.

[0110] The prediction result is whether the sample customers will make a purchase within the next N days.

[0111] It should be noted that the prediction function can be, but is not limited to, the sigmoid function; no restriction is made here.

[0112] In the implementation of this application, if the sigmoid function is used, the feature vector of the fused target training sample text is input into the sigmoid function, and a 1-dimensional probability score with a value between 0 and 1 is obtained, which is the prediction result.

[0113] S404. Determine whether the error between the predicted result and the actual result meets the preset convergence condition.

[0114] Specifically, if the training sample is positive sample text, the true result is that the sample customer will make a purchase within the next N days; if the training sample is negative sample text, the true result is that the sample customer will not make a purchase within the next N days.

[0115] It should be noted that the preset convergence conditions can be set and modified by technical personnel, expert groups, and authorized personnel, and are not limited here.

[0116] Specifically, if it is determined whether the error between the predicted result and the actual result meets the preset convergence condition, then step S405 is executed; if it is determined that the error between the predicted result and the actual result does not meet the preset convergence condition, then step S406 is executed.

[0117] S405. Use neural network models as prediction models.

[0118] S406. Adjust the parameters in the neural network model.

[0119] It should be noted that in the specific implementation of this application, the model can be modified not only by using the preset convergence conditions, but also by setting a certain maximum number of iterations to train the model. This is not limited here.

[0120] Optionally, in another embodiment of this application, one implementation of the method for predicting customer purchase intention further includes:

[0121] Record the predicted purchase intentions of customers today in the intention form.

[0122] The following is an example illustrating the practical application of this application: First, all customer communication data for the day is retrieved from the communication database daily. Then, for each customer, the text of each conversation is concatenated to obtain the customer's communication text for the day. This text is then input into the prediction model. The model first preprocesses the customer's communication text to obtain the preprocessed version. Next, it filters the preprocessed text to obtain the customer's target communication text for the day. Then, it extracts features from the target text to obtain its feature vector. This feature vector is then fused with the feature vectors of the customer's historical target communication texts to obtain the fused feature vector. Finally, the fused feature vector is input into the prediction function, and the predicted result of the customer's purchase intention for the day is output.

[0123] As can be seen from the above scheme, this application provides a method for predicting customer purchase intention: First, all customer communication data for today is read from the communication database; wherein, the communication data includes the text of each conversation with the customer; then, for each customer, the text of each conversation with the customer is concatenated to obtain the customer's communication text for today; then, the customer's communication text for today is input into the prediction model, and the prediction model processes the customer's communication text for today to obtain the prediction result of the customer's purchase intention for today; wherein, the prediction model is obtained by training a neural network model with a training sample set; the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text. This achieves the goal of accurately predicting customer purchase intention.

[0124] Another embodiment of this application provides a device for predicting customer purchase intentions, such as... Figure 6 As shown, it specifically includes:

[0125] Reading unit 601 is used to read all customer communication data for today from the communication database.

[0126] The communication data includes the text of every conversation with the customer.

[0127] The splicing unit 602 is used to splice together the text of each conversation with each customer to obtain the customer's communication text for today.

[0128] The first prediction unit 603 is used to input the customer's communication text of today into the prediction model, and to process the customer's communication text of today through the prediction model to obtain the prediction result of the customer's purchase intention of today.

[0129] The prediction model is obtained by training the neural network model using a training sample set. The training sample set includes at least one training sample text. The training sample text includes positive sample text and negative sample text. A positive sample text is defined as follows: if a customer makes a purchase within N days after engaging in a sample dialogue, the sample dialogue text and the first M samples of the sample dialogue text are considered as positive sample text. A negative sample text is defined as follows: if a customer makes a purchase within N days after engaging in a sample dialogue, the sample dialogue text and the first M samples of the sample dialogue text are considered as negative sample text.

[0130] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 1 As shown, it will not be elaborated further here.

[0131] Optionally, in another embodiment of this application, when the first prediction unit 603 processes the customer's communication text for today using a prediction model to obtain a prediction result of the customer's purchase intention for today, such as... Figure 7 As shown, it includes:

[0132] The first preprocessing unit 701 is used to preprocess the customer's communication text for today to obtain the preprocessed communication text for today.

[0133] The first filtering unit 702 is used to filter the pre-processed customer's communication text for today to obtain the target communication text.

[0134] The first feature extraction unit 703 is used to extract features from the target communication text to obtain the feature vector of the target communication text.

[0135] The fusion unit 704 is used to fuse the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text; wherein, the feature vector of the customer's historical target communication text is obtained by querying the historical feature vector database.

[0136] The first input unit 705 is used to input the feature vector of the fused target communication text into the prediction function to obtain the prediction result of the customer's purchase intention today.

[0137] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 2 As shown, it will not be elaborated further here.

[0138] Optionally, in another embodiment of this application, one implementation of the customer purchase intention prediction device further includes:

[0139] The replacement unit is used to replace the feature vector of the customer's historical target communication text in the historical feature vector database with the feature vector of the fused target communication text.

[0140] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, which will not be repeated here.

[0141] Optionally, in another embodiment of this application, one implementation of the predictive model building unit is as follows: Figure 8 As shown, it includes:

[0142] Training sample set construction unit 801 is used to construct the training sample set.

[0143] The training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; a positive sample text is defined as: if a customer makes a purchase within N days after engaging in a sample dialogue with the customer, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are used as positive sample text; a negative sample text is defined as: if a customer makes a purchase within N days after engaging in a sample dialogue with the customer, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are used as negative sample text.

[0144] The selection unit 802 is used to randomly select a training sample text from the training sample set.

[0145] The second prediction unit 803 is used to input the training sample text into the neural network model, process the training sample text through the neural network model, and output the prediction result.

[0146] The prediction result is whether the sample customers will make a purchase within the next N days.

[0147] Optionally, in another embodiment of this application, the second prediction unit 803 is used in one implementation whereby the training sample text is processed by a neural network model to output a prediction result, such as... Figure 9 As shown, it includes:

[0148] The second preprocessing unit 901 is used to preprocess the training sample text to obtain the preprocessed training sample text.

[0149] The second filtering unit 902 is used to filter the preprocessed training sample text to obtain the target training sample text.

[0150] The second feature extraction unit 903 is used to extract features from the target training sample text to obtain the feature vector of the target training sample text.

[0151] The second input unit 904 is used to input the feature vector of the target training sample text into the prediction function to obtain the prediction result.

[0152] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 5 As shown, it will not be elaborated further here.

[0153] The model determination unit 804 is used to continuously adjust the parameters in the neural network model by utilizing the error between the predicted result and the actual result until the error between the predicted result and the actual result output by the adjusted neural network model meets the preset convergence condition, and then the adjusted neural network model is determined as the prediction model.

[0154] Specifically, if the training sample is positive sample text, the true result is that the sample customer will make a purchase within the next N days; if the training sample is negative sample text, the true result is that the sample customer will not make a purchase within the next N days.

[0155] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 4 As shown, it will not be elaborated further here.

[0156] Optionally, in another embodiment of this application, one implementation of the customer purchase intention prediction device further includes:

[0157] The recording unit is used to record the predicted results of customers' purchase intentions for today into the intention form.

[0158] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, which will not be repeated here.

[0159] As can be seen from the above scheme, this application provides a device for predicting customer purchase intention: First, it reads all customer communication data for today from a communication database; wherein, the communication data includes the text of each conversation with the customer; then, for each customer, it concatenates the text of each conversation with the customer to obtain the customer's communication text for today; then, it inputs the customer's communication text for today into a prediction model, and processes the customer's communication text for today through the prediction model to obtain the prediction result of the customer's purchase intention for today; wherein, the prediction model is obtained by training a neural network model with a training sample set; the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text. This achieves the purpose of accurately predicting customer purchase intention.

[0160] Another embodiment of this application provides an electronic device, such as... Figure 10 As shown, it includes:

[0161] One or more processors 1001.

[0162] Storage device 1002, on which one or more programs are stored.

[0163] When the one or more programs are executed by the one or more processors 1001, the one or more processors 1001 implement the customer purchase intention prediction method as described in any of the above embodiments.

[0164] Another embodiment of this application provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for predicting customer purchase intention as described in any of the above embodiments.

[0165] In the embodiments disclosed in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0166] Furthermore, the functional modules in the various embodiments of this disclosure can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a live streaming device, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0167] Those skilled in the art will be able to implement or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting customer purchase intention, characterized in that, include: Read all customer communication data for today from the communication database; wherein, the communication data includes the text of every conversation with the customer; For each customer, the text of each conversation with the customer is concatenated to obtain the customer's communication text for today; The customer's communication text for today is input into a prediction model. The prediction model processes the customer's communication text for today to obtain a prediction result of the customer's purchase intention for today. The prediction model is trained on a neural network model using a training sample set. The training sample set includes at least one training sample text, which includes positive and negative sample texts. A positive sample text is defined as: if a customer makes a purchase within N days after a sample conversation, the sample conversation text and the first M sample conversation texts of that sample conversation are considered positive sample texts. A negative sample text is defined as: if a customer does not make a purchase within N days after a sample conversation, the sample conversation text and the first M sample conversation texts of that sample conversation are considered negative sample texts. The step of processing the customer's communication text for today using the prediction model to obtain a prediction result of the customer's purchase intention for today includes: The customer's communication text for today is preprocessed to obtain the preprocessed communication text for today. The preprocessed customer communication text for today is filtered to obtain the target communication text; Feature extraction is performed on the target communication text to obtain the feature vector of the target communication text; The feature vector of the target communication text is fused with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text; wherein, the feature vector of the customer's historical target communication text is obtained by querying a historical feature vector database; The feature vector of the fused target communication text is input into the prediction function to obtain the prediction result of the customer's purchase intention today.

2. The prediction method according to claim 1, characterized in that, After fusing the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text, the method further includes: The feature vector of the fused target communication text is used to replace the feature vector of the customer's historical target communication text in the historical feature vector database.

3. The prediction method according to claim 1, characterized in that, The method for constructing the prediction model includes: Construct a training sample set; wherein the training sample set includes: at least one training sample text; the training sample text includes positive sample text and negative sample text; the positive sample text is defined as: if a sample customer makes a purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as positive sample text; the negative sample text is defined as: if a sample customer does not make a purchase within N days after engaging in a sample dialogue text, then the sample dialogue text and the first M sample dialogue texts of the sample dialogue text are considered as negative sample text. In the training sample set, a training sample text is randomly selected; The training sample text is input into a neural network model, which processes the training sample text and outputs a prediction result; wherein, the prediction result is whether the sample customer will make a purchase in the next N days; The parameters in the neural network model are continuously adjusted using the error between the predicted result and the actual result until the error between the predicted result and the actual result output by the adjusted neural network model meets a preset convergence condition. The adjusted neural network model is then determined as the prediction model. Specifically, if the training sample is positive sample text, the actual result is that the sample customer will make a purchase within the next N days; if the training sample is negative sample text, the actual result is that the sample customer will not make a purchase within the next N days.

4. The prediction method according to claim 3, characterized in that, The process of processing the training sample text using the neural network model includes: The training sample text is preprocessed to obtain the preprocessed training sample text. The preprocessed training sample text is filtered to obtain the target training sample text. Feature extraction is performed on the target training sample text to obtain the feature vector of the target training sample text; The feature vector of the target training sample text is input into the prediction function to obtain the prediction result.

5. The prediction method according to claim 1, characterized in that, After outputting the predicted purchase intention of the customer for today, the output also includes: Record the predicted purchase intentions of the customers today in the intention table.

6. A device for predicting customer purchase intention, characterized in that, include: The reading unit is used to read all customer communication data for today from the communication database; wherein, the communication data includes the text of each conversation with the customer; The splicing unit is used to splice together the text of each conversation with the customer for each customer to obtain the customer's communication text for today. The first prediction unit is used to input the customer's communication text for today into the prediction model, and process the customer's communication text for today through the prediction model to obtain the prediction result of the customer's purchase intention for today. The prediction model is obtained by training a neural network model using a training sample set. The training sample set includes at least one training sample text; the training sample text includes positive sample text and negative sample text; the positive sample text is defined as: if a purchase occurs within N days after a sample conversation with the sample customer, then the sample conversation text and the first M sample conversation texts of the sample conversation text are considered positive sample text; the negative sample text is defined as: if no purchase occurs within N days after a sample conversation with the sample customer, then the sample conversation text and the first M sample conversation texts of the sample conversation text are considered negative sample text. Wherein, the first prediction unit is used to process the customer's communication text for today through the prediction model to obtain the prediction result of the customer's purchase intention for today, including: The first preprocessing unit is used to preprocess the customer's communication text for today to obtain the preprocessed communication text for today. The first filtering unit is used to filter the preprocessed customer's communication text of the day to obtain the target communication text; The first feature extraction unit is used to extract features from the target communication text to obtain the feature vector of the target communication text; The fusion unit is used to fuse the feature vector of the target communication text with the feature vector of the customer's historical target communication text to obtain the feature vector of the fused target communication text; wherein, the feature vector of the customer's historical target communication text is obtained by querying a historical feature vector database; The first input unit is used to input the feature vector of the fused target communication text into the prediction function to obtain the prediction result of the customer's purchase intention today.

7. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the customer purchase intention prediction method as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the method for predicting customer purchase intention as described in any one of claims 1 to 5.

Images