Marketing effect quality inspection method, marketing effect quality inspection device, marketing effect quality inspection equipment and medium

Abstract

The invention relates to the field of finance, in particular to a marketing effect quality inspection method and device, equipment and a medium, and the method comprises the steps: obtaining a historical text of a marketing call and a corresponding historical label; obtaining a historical feature extraction result corresponding to the historical text by using a first encoder and a second encoder; performing classification prediction on the historical feature extraction result by using a classifier to obtain historical prediction labels, and obtaining a trained first encoder, a trained second encoder and a trained classifier according to the historical labels and the historical prediction labels; a current text to be subjected to quality inspection is obtained, the current text sequentially passes through a trained first encoder, a trained second encoder and a trained classifier to classify a current feature extraction result to obtain a quality inspection result, feature encoding and context feature extraction are performed on the text of the marketing call, so that classification of the real-time text of the marketing call is more accurate, and the quality inspection accuracy is improved. Therefore, a more effective marketing strategy of financial business can be formulated and adjusted in time.

Description

Technical Field

[0001] The present invention relates to the financial field, and in particular to a marketing effect quality inspection method, device, equipment and medium. Background Art

[0002] Marketing effect is an important evaluation indicator for enterprises when conducting telephone marketing. Whether an enterprise can accurately understand the marketing effect of business opportunities plays a key role in the marketing performance of the enterprise and is related to the development of the enterprise. For example, in enterprises such as banks, financial institutions and insurance institutions, salesmen mostly sell their products through telephone marketing or telephone return visits. Efficient and accurate quality inspection of the marketing effect of business opportunities can quickly understand the marketing effect, formulate and adjust more effective marketing strategies in a timely manner, and help the sales team achieve sales goals, increase product sales, and help enterprises achieve greater profit goals.

[0003] The current marketing effect of business opportunities mainly relies on manual spot checks of business opportunity marketing call records or keyword matching of call texts to evaluate the marketing effect of sales personnel. However, firstly, due to the huge number of business opportunity marketing calls, even spot checks of some marketing call records require a lot of manpower, and the manual spot check method is subjective. Secondly, the keyword matching method of call texts can only cover a few scenarios and has a high error rate. Therefore, an intelligent quality inspection method for marketing effects is needed to improve quality inspection efficiency. Summary of the invention

[0004] In view of this, the embodiments of the present invention provide a marketing effect quality inspection method, device, equipment and medium to solve the problem of low efficiency of marketing effect quality inspection.

[0005] In a first aspect, an embodiment of the present invention provides a marketing effect quality inspection method, the marketing effect quality inspection method comprising: Obtain historical texts of marketing calls within a historical time period and corresponding historical marketing results, and use the historical marketing results as historical tags of the historical texts; Using a first encoder to perform feature encoding on the historical text to obtain text vectorization features, and using a second encoder to perform context feature extraction on the text vectorization features to obtain a historical feature extraction result corresponding to the historical text; Using a classifier to perform classification prediction on the historical feature extraction results to obtain historical classification results; The historical classification results are used as historical prediction labels, and according to the historical labels and the historical prediction labels, the parameters of the first encoder, the second encoder, and the classifier are adjusted and the training is repeated until an iteration condition is satisfied, thereby obtaining the trained first encoder, the second encoder, and the classifier; Acquire a current text to be quality-checked, and perform feature extraction on the current text by using a trained first encoder and a trained second encoder in sequence to obtain a current feature extraction result; The current feature extraction result is classified by a trained classifier to obtain a current classification result, and the current classification result is determined to be a quality inspection result of the current text.

[0006] In a second aspect, an embodiment of the present invention provides a marketing effect quality inspection device, the marketing effect quality inspection device comprising: A historical data acquisition module, used to acquire historical texts of marketing calls and corresponding historical marketing results within a historical time period, and use the historical marketing results as historical tags of the historical texts; A first feature extraction module is used to perform feature encoding on the historical text using a first encoder to obtain text vectorization features, and to perform context feature extraction on the text vectorization features using a second encoder to obtain a historical feature extraction result corresponding to the historical text; A label prediction module, used to use a classifier to perform classification prediction on the historical feature extraction results to obtain historical classification results; A parameter adjustment module, used to use the historical classification results as historical prediction labels, and adjust the parameters of the first encoder, the second encoder, and the classifier according to the historical labels and the historical prediction labels, and repeat the training until an iteration condition is met, thereby obtaining the trained first encoder, the second encoder, and the classifier; A second feature extraction module is used to obtain the current text to be inspected, and extract features from the current text through the trained first encoder and the trained second encoder in sequence to obtain a current feature extraction result; The quality inspection result acquisition module is used to classify the current feature extraction result through a trained classifier to obtain a current classification result, and determine that the current classification result is the quality inspection result of the current text.

[0007] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the marketing effectiveness quality inspection method as described in the first aspect when executing the computer program.

[0008] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the marketing effectiveness quality inspection method as described in the first aspect is implemented.

[0009] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: The present invention obtains the historical text of marketing calls within the historical time period and the corresponding historical marketing results, and uses the historical marketing results as the historical labels of the historical texts; uses the first encoder to perform feature encoding on the historical texts to obtain text vectorization features, and uses the second encoder to perform context feature extraction on the text vectorization features to obtain the historical feature extraction results of the corresponding historical texts; uses a classifier to perform classification and prediction on the historical feature extraction results to obtain historical classification results, and uses the historical classification results as historical prediction labels. According to the historical labels and historical prediction labels, the parameters of the first encoder, the second encoder and the classifier are adjusted and the training is repeated until the iteration conditions are met, so as to obtain the trained first encoder, the second encoder and the classifier; obtains the current The text is extracted by the trained first encoder and the trained second encoder in turn to obtain the current feature extraction result, the current feature extraction result is classified by the trained classifier to obtain the current classification result, and the current classification result is determined to be the quality inspection result of the current text. By performing feature encoding and context feature extraction on historical texts, the text information contained in the acquired feature extraction result is more comprehensive and more conducive to subsequent text classification, thereby improving the ability of the trained first encoder, second encoder and classifier obtained by training according to the historical classification results, making the classification of real-time text of marketing calls more accurate, thereby facilitating the timely formulation and adjustment of more effective marketing strategies for financial services. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] Figure 1 This is a schematic diagram of an application environment of a marketing effect quality inspection method provided by the first embodiment of the present invention; Figure 2 It is a flowchart of a marketing effect quality inspection method provided in the first embodiment of the present invention; Figure 3 It is a structural schematic diagram of a marketing effect quality inspection device provided in Embodiment 2 of the present invention; Figure 4 It is a structural diagram of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0012] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0013] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0014] It should also be understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0015] As used in the present specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]", depending on the context.

[0016] In addition, in the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0017] References to "one embodiment" or "some embodiments" etc. described in the present specification mean that one or more embodiments of the present invention include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0018] It should be understood that the order of execution of the steps in the following embodiments does not imply a precedence of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0019] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.

[0020] A marketing effect quality inspection method provided in the first embodiment of the present invention can be applied in the following aspects: Figure 1 In the application environment, the client communicates with the server. The client includes but is not limited to PDAs, desktop computers, laptops, ultra-mobile personal computers (UMPCs), netbooks, cloud terminal devices, personal digital assistants (PDAs) and other computer devices. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0021] The marketing effectiveness quality inspection method is applied in a financial scenario. The client may be a mobile phone. The mobile phone connects to a server via a wireless network to send marketing call texts when a salesperson uses a mobile phone for telephone marketing or telephone return visits to sell financial business products. The marketing effect quality inspection of the marketing call texts is then performed on the server to timely understand the marketing effect.

[0022] See also Figure 2 , is a flow chart of a marketing effect quality inspection method provided by Embodiment 1 of the present invention. The marketing effect quality inspection method can be applied to Figure 1 The server in the example, the computer device corresponding to the server is connected to the corresponding database to implement the quality inspection of the marketing effect based on the historical marketing call texts stored in the database. In addition, the server is also connected to the client to obtain the marketing call texts sent by the user from the client. Figure 2 As shown, the marketing effect quality inspection method may include the following steps: Step S201, obtaining historical texts of marketing calls within a historical time period and corresponding historical marketing results, and using the historical marketing results as historical tags of the historical texts.

[0023] The historical time period can be one or more hours, or one or more days, or one or more weeks, etc. Assuming it is 1 hour and the time starts at 12 o'clock, the historical time periods are 11 o'clock, 10 o'clock, 9 o'clock, etc. Marketing calls refer to the call recording data between the salesperson and the customer, and historical texts refer to the conversation texts between the salesperson and the customer saved in the historical time period. Marketing effects include effective marketing follow-ups and ineffective marketing follow-ups. Tags are used to reflect the marketing effects corresponding to the texts of marketing calls. The corresponding historical tags can be manually annotated. Therefore, the historical texts of marketing calls and the corresponding historical marketing results within the historical time period are obtained, and the historical marketing results are used as historical tags for historical texts.

[0024] For example, in an insurance sales scenario, the marketing effects are divided into two categories, namely effective marketing follow-up and ineffective marketing follow-up, with corresponding label values ​​of 1 and 0. The historical text of the marketing call corresponding to the salesperson selling insurance to the customer over the phone is obtained. When the customer buys the insurance, it means that the salesperson has successfully sold the insurance, and the historical label of the corresponding historical text is 1. Conversely, when the customer does not buy the insurance, it means that the salesperson has not made a successful sale, and the historical label of the corresponding historical text is 0.

[0025] Step S202: Use the first encoder to perform feature encoding on the historical text to obtain text vectorization features, and use the second encoder to perform context feature extraction on the text vectorization features to obtain historical feature extraction results corresponding to the historical text.

[0026] Among them, the first encoder and the second encoder are used to extract features from the corresponding input samples, and the first encoder and the second encoder have certain applicability and good flexibility in selection. Therefore, the first encoder is used to perform feature encoding on the historical text to obtain text vectorization features, and the second encoder is used to perform contextual feature extraction on the text vectorization features to obtain the historical feature extraction results of the corresponding historical text, so that the historical feature extraction results obtained can pay attention to local information and effectively take into account global information.

[0027] Optionally, the first encoder is used to perform feature encoding on the historical text to obtain text vectorization features, including: Perform word segmentation processing on the historical text to obtain the word segmentation result; Get the word vector of each word in the word segmentation result and the position code of each word in the dictionary, serialize the word segmentation result according to the word vector and position code, and obtain the text vectorization features of the historical text.

[0028] Among them, the first encoder can be an ALBERT (A lite BERT) model, which includes an input layer, a coding layer and an output layer. The model trains deep bidirectional representation by combining bidirectional converters in all layers, and the model input vector of the model is the sum of the vectors of word vectors, position vectors and sentence vectors, wherein the word vector is a vectorized representation of the text, the position vector is used to represent the position of the word in the text, and the sentence vector is used to represent the order of sentences in the text. Therefore, using the ALBERT model to perform feature encoding on historical texts can reduce calculation parameters and speed up training.

[0029] The purpose of preprocessing historical texts is to extract useful data from historical texts, or to delete noise data from historical texts, so that data in historical texts that are irrelevant to the purpose of extraction can be deleted. The word segmentation results are input into the ALBERT model, and the historical texts are serialized to obtain the corresponding text vectorization features. Specifically, for each word in the word segmentation results corresponding to the historical text, the part of speech of each word is first converted into a part-of-speech vector, and then the positional encoding is used to add position information to each word in the word segmentation results, and the initial word vector is used to represent the historical text with added position information. Finally, the initial word vector and the part-of-speech vector are added to obtain the word vector sequence corresponding to the historical text, and the word vector sequence is used as the text vectorization feature of the historical text.

[0030] For example, in an insurance sales scenario, the historical text of the corresponding marketing calls when the salesperson sells insurance to customers over the phone is obtained, the historical text is preprocessed to obtain the corresponding word segmentation results, and the word segmentation results are input into the ALBERT model to perform feature encoding on the historical text to obtain the corresponding text vectorization features.

[0031] Optionally, preprocess the historical text to obtain word segmentation results, including: Stop words are removed from the historical text to obtain a first historical text, garbled characters are removed from the first historical text to obtain a second historical text, text error correction is performed on the second historical text to obtain a third historical text, and a word segmentation operation is performed on the third historical text to obtain a word segmentation result.

[0032] Among them, stop words refer to certain characters or words that are automatically filtered out before or after processing natural speech text in order to save storage space and improve search efficiency in information retrieval. These characters or words are called stop words. In an embodiment, the words that do not contribute to the text features in the natural language text can be removed by removing stop words, such as punctuation marks, personal names, spaces, etc. Then the method of removing stop words from the historical text to obtain the first historical text is: stop word list filtering, stop word list filtering refers to matching the words in the text with the already constructed stop word list one by one. If the match is successful, the corresponding word is a stop word and needs to be deleted.

[0033] The method for removing garbled characters from the first historical text and obtaining the second historical text is: garbled character table filtering. Garbled character table filtering refers to matching each character in the historical text with the constructed garbled character table. If the matches are the same, the corresponding character is garbled character and needs to be deleted.

[0034] Text correction refers to the use of machine learning and natural speech processing technology to automatically correct text information. Therefore, text correction is performed on the second historical text to obtain a third historical text. The specific text correction can refer to the existing correction method, which will not be repeated in the embodiment of this method.

[0035] In order to obtain the historical text in vector form, it is necessary to segment the historical text. Segmentation is a basic task in lexical analysis. Segmentation algorithms are mainly divided into two categories according to their core ideas: one is dictionary-based segmentation, which first divides the text data into words according to the dictionary, and then finds the best combination of words; the other is character-based segmentation, which is based on the word formation, first divides the sentence into individual words, and then combines the words into words to find the optimal segmentation strategy. In this embodiment, a statistical-based segmentation method is used to perform a segmentation operation on the third historical text to obtain a segmentation result.

[0036] Based on the statistical word segmentation method, from a formal point of view, a word is a stable combination of characters. In the context, the more times adjacent characters appear at the same time, the more likely they are to form a word. Therefore, the frequency or probability of adjacent co-occurrence of characters can better reflect the credibility of the word, that is, its importance in the text. By counting the frequency of the combination of adjacent co-occurring characters in the historical text, the closeness of the combination relationship between them is calculated. When the closeness is higher than the preset threshold, it is confirmed that this character group can form a word, and then the corresponding word segmentation result of the historical text is obtained.

[0037] For example, in an insurance sales scenario, the historical text of the marketing calls corresponding to the salesperson selling insurance to the customer over the phone is obtained, and the words in the historical text are matched with the constructed stop word list. If the match is successful, the corresponding word is a stop word and needs to be deleted to obtain the first historical text; the constructed garbled code table is used to match each character in the historical text. If the match is the same, the corresponding character is garbled code and needs to be deleted to obtain the second historical text; the second historical text is corrected using the existing error correction method to obtain the third historical text; finally, the third historical text is segmented using a statistical-based word segmentation method to obtain a word segmentation result.

[0038] Optionally, a second encoder is used to perform context feature extraction on the text vectorized features to obtain a historical feature extraction result corresponding to the historical text, including: The text vectorized features are input into the second encoder and processed in the positive time order to obtain the forward implicit output vector; The text vectorized features are input into the second encoder and processed in reverse time order to obtain a reverse implicit output vector; The forward implicit output vector and the reverse implicit output vector are concatenated according to the position to obtain the historical feature extraction result of the corresponding historical text. Among them, the second encoder can be a bidirectional recurrent neural network (BiGRU, Bidirectional Recurrent Neural Network), which is a variant of the long short-term memory network (LSTM, Long Short-Term Memory). It combines the forget gate and the input gate into an update gate, which has a simpler structure, fewer parameters and reduces the model time cost. The BiGRU neural network is a neural network composed of a forward BiGRU, a backward BiGRU, an output state connection layer of the forward BiGRU, and an output state connection layer of the backward BiGRU. BiGRU can learn relatively global functions and reduce the required parameters.

[0039] For the application of BiGRU neural network, the text sequence is input into the forward BiGRU from both the forward and backward directions, and the hidden state outputs of each time step in the hidden layer are spliced ​​as the input of each time step of the backward BiGRU. The backward BiGRU also splices and outputs the hidden state of each time step in the hidden layer. Therefore, the text vectorization features are input into the BiGRU neural network, and the forward BiGRU and the backward BiGRU read them in the forward order (time positive order) and the reverse order (time reverse order) respectively, and the corresponding forward implicit output vector and reverse implicit output vector are output. Then the forward implicit output vector The expression is ,in, is the output of the forward BiGRU at the tth time step, is the output of the forward BiGRU at the t-1th time step, is the input of the forward BiGRU at the tth time step, and the reverse implicit output vector The expression is ,in, is the output of the backward BiGRU at the tth time step, is the output of the t-1th time step of the backward BiGRU, is the input of the backward BiGRU at the tth time step; then the forward implicit output vector and the reverse implicit output vector are concatenated in order of position to obtain the complete implicit feature vector, which is the output of the BiGRU neural network. The complete implicit feature vector The expression is: , thereby confirming that the complete implicit feature vector is the historical feature extraction result of the corresponding historical text. The BiGRU neural network helps to extract deeper text context semantic information and dependencies. The outputs of the two BiGRUs before and after are matched with the input text sequence in time step order and spliced. This avoids the occurrence of gradient diffusion or explosion during training, and at the same time makes the semantic information contained in the output vector richer, avoiding the loss of important information to a certain extent.

[0040] For example, in an insurance sales scenario, the ALBERT model is used to obtain the text vectorization features corresponding to the historical text of the marketing calls corresponding to the salesperson selling insurance to customers over the phone, and then the text vectorization features are input into the BiGRU neural network, and read by the forward BiGRU and the backward BiGRU in forward order and reverse order respectively, and the forward implicit output vector and the reverse implicit output vector are output accordingly. The forward implicit output vector and the reverse implicit output vector are concatenated in position order to obtain a complete implicit feature vector, which is the historical feature extraction result of the corresponding historical text.

[0041] Step S203, use a classifier to perform classification prediction on the historical feature extraction results to obtain historical classification results, use the historical classification results as historical prediction labels, and adjust the parameters of the first encoder, the second encoder, and the classifier according to the historical labels and the historical prediction labels, and repeat the training until the iteration conditions are met to obtain the trained first encoder, the second encoder, and the classifier.

[0042] Among them, the classification result refers to the marketing effect category to which the text belongs, the trained first encoder refers to the first encoder that has completed training and has a strong local feature extraction capability, and similarly, the trained second encoder refers to the second encoder that has completed training and also has a strong global feature extraction capability, and the trained classifier refers to the trained classifier that has the ability to classify the object to be identified. In some embodiments, the trained first encoder, second encoder and classifier are obtained by training based on large sample data. Therefore, the classifier is used to classify and predict the historical feature extraction results to obtain the historical classification results, and the historical classification results are used as historical prediction labels. Then, a supervised training method is adopted, and according to the loss between the historical label and the historical prediction label, the parameters of the first encoder, the second encoder and the classifier are adjusted and the training is repeated until the iteration condition is met, so as to obtain the trained first encoder, the second encoder and the classifier. Among them, the specific calculation of the loss between the historical label and the historical prediction label can refer to the existing method, which will not be repeated in the embodiment of this method.

[0043] Preferably, this embodiment uses the cross entropy loss function to calculate the loss between the historical labels and the historical prediction labels, and then reversely corrects the parameters of the first encoder, the second encoder and the classifier according to the gradient descent method until the loss between the historical labels and the historical prediction labels converges, thereby obtaining the trained first encoder, the second encoder and the classifier.

[0044] Optionally, a classifier is used to perform classification prediction on the historical feature extraction results to obtain historical classification results, including: Input the historical feature extraction results into the classifier to obtain the classification prediction results; If the classification prediction result is greater than the preset classification result threshold, the historical classification result is confirmed as an effective marketing follow-up; If the classification prediction result is not greater than the preset classification result threshold, the historical classification result is confirmed as invalid marketing follow-up.

[0045] Among them, the classification prediction result can refer to the classification prediction value output by the classifier, which is used to characterize the probability of the marketing effect category to which the text belongs, and the corresponding preset classification result threshold refers to the threshold of the classification prediction value. The classifier can be a random forest classifier, which is composed of multiple decision trees, and each decision tree includes multiple nodes, and the multiple nodes have a hierarchical relationship. For each node, the node can be connected to a node in the previous layer as a branch node of the previous layer node, and the node can be connected to at least one branch node in the next layer. The nodes except the last layer node have split dimensions and split conditions, and the last layer node has a corresponding classification result. Therefore, the historical feature extraction result is input into the random forest classifier to obtain the corresponding classification prediction value, and then the classification prediction value is compared with the preset classification prediction threshold. If the classification prediction result is greater than the preset classification result threshold, it is considered that the marketing call corresponding to the corresponding historical text is effective marketing, and the historical classification result is confirmed to be effective marketing follow-up. On the contrary, if the classification prediction result is not greater than the preset classification result threshold, it is considered that the marketing call corresponding to the corresponding historical text is invalid marketing, and the historical classification result is confirmed to be invalid marketing follow-up. Among them, the preset classification result threshold is 0.5.

[0046] For example, in an insurance sales scenario, after feature extraction is performed on the historical text of the marketing calls corresponding to the salesperson selling insurance to the customer over the phone and the corresponding historical feature extraction results are obtained, the historical feature extraction results are input into a random forest classifier to obtain the corresponding classification prediction value, and then the classification prediction value is compared with a preset classification prediction threshold to obtain a comparison result, so as to determine the historical classification result based on the comparison result.

[0047] Step S204, obtain the current text to be quality inspected, extract features of the current text by the trained first encoder and the trained second encoder in turn, and obtain the current feature extraction result; classify the current feature extraction result by the trained classifier to obtain the current classification result, and determine the current classification result as the quality inspection result of the current text.

[0048] The current text to be quality inspected refers to the marketing call text that has not been quality inspected for marketing effects. The trained first encoder and the second encoder have the ability to accurately and comprehensively extract text features from the text to be quality inspected except for the sample data. Therefore, the current text to be quality inspected is obtained, and the current text is sequentially subjected to feature extraction by the trained first encoder and the second encoder to obtain the current feature extraction result; and the trained classifier also has the ability to accurately classify the feature extraction results output by the first encoder and the second encoder. Therefore, the trained classifier is used to classify the current feature extraction result to obtain the current classification result, and the current classification result is determined to be the quality inspection result of the current text.

[0049] In summary, the embodiments of the present invention obtain the historical text of marketing calls within the historical time period and the corresponding historical marketing results, and use the historical marketing results as the historical labels of the historical texts; use the first encoder to perform feature encoding on the historical texts to obtain text vectorization features, and use the second encoder to perform context feature extraction on the text vectorization features to obtain the historical feature extraction results of the corresponding historical texts; use the classifier to perform classification and prediction on the historical feature extraction results to obtain historical classification results, and use the historical classification results as historical prediction labels, and adjust the parameters of the first encoder, the second encoder, and the classifier according to the historical labels and the historical prediction labels, and repeat the training until the iteration conditions are met, so as to obtain the trained first encoder, the second encoder, and the classifier; obtain For the current text to be quality inspected, the current text is successively subjected to feature extraction by the trained first encoder and the second encoder to obtain the current feature extraction result, the current feature extraction result is classified by the trained classifier to obtain the current classification result, and the current classification result is determined to be the quality inspection result of the current text. By performing feature encoding and context feature extraction on historical texts, the text information contained in the acquired feature extraction result is more comprehensive and more conducive to subsequent text classification, thereby improving the ability of the trained first encoder, the second encoder and the classifier obtained by training according to the historical classification results, making the classification of the real-time text of the marketing call more accurate, thereby facilitating the timely formulation and adjustment of more effective marketing strategies for financial services.

[0050] Corresponding to the marketing effect quality inspection method in the above embodiment, Figure 3 FIG. 1 shows a structural block diagram of a marketing effect quality inspection device provided by Embodiment 2 of the present invention, see Figure 3 , the marketing effect quality inspection device comprises: A historical data acquisition module 31 is used to acquire historical texts of marketing calls and corresponding historical marketing results within a historical time period, and use the historical marketing results as historical tags of the historical texts; A first feature extraction module 32 is used to perform feature encoding on the historical text using a first encoder to obtain text vectorization features, and to perform context feature extraction on the text vectorization features using a second encoder to obtain a historical feature extraction result corresponding to the historical text; The label prediction module 33 is used to use a classifier to perform classification prediction on the historical feature extraction result to obtain a historical classification result; A parameter adjustment module 34 is used to use the historical classification result as a historical prediction label, and according to the historical label and the historical prediction label, adjust the parameters of the first encoder, the second encoder and the classifier, and repeat the training until the iteration condition is met, so as to obtain the trained first encoder, the second encoder and the classifier; The second feature extraction module 35 is used to obtain the current text to be inspected, and extract features from the current text through the trained first encoder and the trained second encoder in sequence to obtain a current feature extraction result; The quality inspection result acquisition module 36 is used to classify the current feature extraction result through a trained classifier to obtain a current classification result, and determine that the current classification result is the quality inspection result of the current text.

[0051] Optionally, the first feature extraction module 32 includes: A preprocessing unit is used to perform word segmentation processing on the historical text to obtain the word segmentation result; A word segmentation processing unit, used to obtain the word vector of each word in the word segmentation result and the corresponding position code of each word in the dictionary; The serialization processing unit is used to serialize the word segmentation results according to the word vector and position encoding to obtain the text vectorization features of the historical text.

[0052] Optionally, the first feature extraction module 32 includes: A forward order extraction unit, used for inputting the text vectorized features into the second encoder, processing them in the forward order of time, and obtaining a forward implicit output vector; A reverse extraction unit, used for inputting the text vectorized features into the second encoder, processing them in reverse time order, and obtaining a reverse implicit output vector; The vector concatenation unit is used to concatenate the forward implicit output vector and the reverse implicit output vector according to the position to obtain the historical feature extraction result of the corresponding historical text.

[0053] Optionally, the quality inspection result acquisition module 36 includes: A result prediction unit, used to input the historical feature extraction results into the classifier to obtain the classification prediction results; A first confirmation unit, used to confirm that the historical classification result is an effective marketing follow-up if the classification prediction result is greater than a preset classification result threshold; The second confirmation unit is used to confirm that the historical classification result is an invalid marketing follow-up if the classification prediction result is not greater than a preset classification result threshold.

[0054] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules and units are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0055] Figure 4 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Figure 4 As shown, the computer device of this embodiment includes: at least one processor ( Figure 4 Only one is shown), a memory, and a computer program stored in the memory and executable on at least one processor, and when the processor executes the computer program, the steps in any of the above-mentioned marketing effectiveness quality inspection method embodiments are implemented.

[0056] The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that Figure 4 This is merely an example of a computer device and does not constitute a limitation on the computer device. The computer device may include more or fewer components than those shown in the figure, or a combination of certain components, or different components. For example, it may also include a network interface, a display screen, and an input device.

[0057] The processor may be a CPU, or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0058] The memory includes a readable storage medium, an internal memory, etc., wherein the internal memory may be the memory of a computer device, and the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage medium. The readable storage medium may be a hard disk of a computer device, and in other embodiments, it may also be an external storage device of the computer device, for example, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device. Further, the memory may also include both an internal storage unit of the computer device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of a computer program, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0059] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the above-mentioned method embodiment, which will not be repeated here. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0060] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by a computer program product. When the computer program product runs on a computer device, the computer device can implement the steps in the above-mentioned method embodiment when executing.

[0061] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0062] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0063] In the embodiments provided by the present invention, it should be understood that the disclosed devices / computer equipment and methods can be implemented in other ways. For example, the device / computer equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0064] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A marketing effect quality inspection method, characterized in that: include: Obtain historical texts of marketing calls within a historical time period and corresponding historical marketing results, and use the historical marketing results as historical tags of the historical texts; Using a first encoder to perform feature encoding on the historical text to obtain text vectorization features, and using a second encoder to perform context feature extraction on the text vectorization features to obtain a historical feature extraction result corresponding to the historical text; Using a classifier to perform classification prediction on the historical feature extraction results to obtain historical classification results; The historical classification results are used as historical prediction labels, and according to the historical labels and the historical prediction labels, the parameters of the first encoder, the second encoder, and the classifier are adjusted and the training is repeated until an iteration condition is satisfied, thereby obtaining the trained first encoder, the second encoder, and the classifier; Acquire a current text to be quality-checked, and perform feature extraction on the current text by using a trained first encoder and a trained second encoder in sequence to obtain a current feature extraction result; The current feature extraction result is classified by a trained classifier to obtain a current classification result, and the current classification result is determined to be a quality inspection result of the current text.

2. The marketing effect quality inspection method according to claim 1, characterized in that: The using the first encoder to perform feature encoding on the historical text to obtain text vectorization features includes: Performing word segmentation processing on the historical text to obtain a word segmentation result; Obtaining the word vector of each word in the word segmentation result and the position code of each word in the dictionary; The word segmentation result is serialized according to the word vector and the position code to obtain text vectorization features of the historical text.

3. The marketing effect quality inspection method according to claim 1, characterized in that: The using the second encoder to perform context feature extraction on the text vectorization feature to obtain a historical feature extraction result corresponding to the historical text includes: Inputting the text vectorization feature into the second encoder, processing it in chronological order, and obtaining a forward implicit output vector; Inputting the text vectorization feature into the second encoder, processing it in reverse time order, and obtaining a reverse implicit output vector; The forward implicit output vector and the reverse implicit output vector are spliced ​​according to positions to obtain a historical feature extraction result corresponding to the historical text.

4. The marketing effect quality inspection method according to claim 1, characterized in that: The using a classifier to perform classification prediction on the historical feature extraction result to obtain a historical classification result includes: Inputting the historical feature extraction result into the classifier to obtain a classification prediction result; If the classification prediction result is greater than a preset classification result threshold, confirm that the historical classification result is an effective marketing follow-up; If the classification prediction result is not greater than a preset classification result threshold, the historical classification result is confirmed to be an invalid marketing follow-up.

5. A marketing effect quality inspection device, characterized in that: include: A historical data acquisition module, used to acquire historical texts of marketing calls and corresponding historical marketing results within a historical time period, and use the historical marketing results as historical tags of the historical texts; A first feature extraction module is used to perform feature encoding on the historical text using a first encoder to obtain text vectorization features, and to perform context feature extraction on the text vectorization features using a second encoder to obtain a historical feature extraction result corresponding to the historical text; A label prediction module, used to use a classifier to perform classification prediction on the historical feature extraction results to obtain historical classification results; A parameter adjustment module, used to use the historical classification results as historical prediction labels, and adjust the parameters of the first encoder, the second encoder, and the classifier according to the historical labels and the historical prediction labels, and repeat the training until an iteration condition is met, thereby obtaining the trained first encoder, the second encoder, and the classifier; A second feature extraction module is used to obtain the current text to be inspected, and extract features from the current text through the trained first encoder and the trained second encoder in sequence to obtain a current feature extraction result; The quality inspection result acquisition module is used to classify the current feature extraction result through a trained classifier to obtain a current classification result, and determine that the current classification result is the quality inspection result of the current text.

6. The marketing effect quality inspection device according to claim 5, characterized in that: The first feature extraction module comprises: A preprocessing unit, used for performing word segmentation processing on the historical text to obtain a word segmentation result; A word segmentation processing unit, used to obtain a word vector of each word in the word segmentation result and a corresponding position code of each word in the dictionary; A serialization processing unit is used to perform serialization processing on the word segmentation result according to the word vector and the position code to obtain the text vectorization feature of the historical text.

7. The marketing effect quality inspection device according to claim 5, characterized in that: The first feature extraction module comprises: A positive order extraction unit, used for inputting the text vectorization feature into the second encoder, processing it in a time positive order, and obtaining a forward implicit output vector; A reverse extraction unit, used for inputting the text vectorization feature into the second encoder, processing it in reverse time order, and obtaining a reverse implicit output vector; The vector concatenation unit is used to concatenate the forward implicit output vector and the reverse implicit output vector according to their positions to obtain a historical feature extraction result corresponding to the historical text.

8. The marketing effect quality inspection device according to claim 5, characterized in that: The quality inspection result acquisition module includes: A result prediction unit, used for inputting the historical feature extraction result into the classifier to obtain a classification prediction result; A first confirmation unit, configured to confirm that the historical classification result is an effective marketing follow-up if the classification prediction result is greater than a preset classification result threshold; The second confirmation unit is used to confirm that the historical classification result is an invalid marketing follow-up if the classification prediction result is not greater than a preset classification result threshold.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the marketing effectiveness quality inspection method according to any one of claims 1 to 5 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the marketing effectiveness quality inspection method as described in any one of claims 1 to 5 is implemented.

Images