Franchise store operation optimization method and system based on big data

Abstract

The present application relates to the field of operation optimization technology, and specifically discloses a franchise store operation optimization method and system based on big data, which first extracts a set of evaluation data for franchise stores from customer feedback data, and uses data processing technology based on deep learning to semantically encode each evaluation data, extracts the semantic embedding feature representation of each evaluation data, and then, through multi-scale temporal context semantic association analysis of each evaluation data, to dig out the implicit customer emotion change trend in the evaluation data, so as to realize intelligent recognition of customer emotion polarity change. In this way, the needs and preferences of customers can be grasped more accurately, so as to adjust the menu, promotion strategy and business strategy to improve customer satisfaction and loyalty, and then achieve better business benefits.

Description

Technical Field

[0001] The present application relates to the field of operation optimization technology, and more specifically, to a franchise store operation optimization method and system based on big data. Background Art

[0002] In today's business environment, franchise stores are a common business model that is widely favored because they can quickly expand brand influence and market share. However, with the increase in the number of franchise stores, how to effectively manage and optimize the operations of each franchise store has become an important issue that brands need to solve urgently. Traditional operation management methods often rely on manual experience and regular market research, which is not only inefficient, but also difficult to fully and timely capture market dynamics and changes in customer needs.

[0003] In recent years, with the rapid development of big data technology, more and more companies have begun to try to use big data for operational decision-making. Big data has the characteristics of large data volume, diverse types, and fast processing speed, which can provide companies with more comprehensive and in-depth insights. In the field of franchise store operations, the application of big data technology also has great potential.

[0004] In the current franchise store operations, customer feedback is an important basis for evaluating operational effectiveness and improving service quality. However, traditional customer feedback analysis methods are usually limited to keyword queries and simple classification statistics of each evaluation data, lacking in-depth exploration and understanding of customer emotions. In addition, customer emotions towards dishes may change with time, season, store promotions and other factors. Therefore, real-time analysis of customer emotion trends can help franchise stores adjust menus or promotion strategies in a timely manner. However, traditional customer evaluation analysis methods usually only focus on the static analysis of a single evaluation data, ignoring the time series information and contextual associations in the evaluation data, making it difficult to capture the dynamic changes in customer emotions, which in turn affects the accuracy and timeliness of operational decisions.

[0005] Therefore, an optimized franchise store operation optimization method based on big data is needed to solve the above technical problems. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a franchise store operation optimization method and system based on big data, which first extracts a set of evaluation data for franchise stores from customer feedback data, and uses deep learning-based data processing technology to semantically encode each evaluation data, extract the semantic embedding feature representation of each evaluation data, and then perform multi-scale temporal context semantic association analysis on each evaluation data to dig out the implicit customer emotion change trend in the evaluation data, thereby realizing intelligent identification of customer emotion polarity changes. In this way, customer needs and preferences can be grasped more accurately, thereby adjusting menus, promotion strategies and business strategies to improve customer satisfaction and loyalty, and thus achieve better operating benefits.

[0007] According to one aspect of the present application, a franchise store operation optimization method based on big data is provided, which includes:

[0008] Extracting a collection of evaluation data for franchise stores from customer feedback data;

[0009] Performing a multi-scale temporal context analysis on the set of evaluation data of the franchise stores to obtain a time sequence of a first time-scale evaluation data context semantic association implicit representation vector and a time sequence of a second time-scale evaluation data context semantic association implicit representation vector;

[0010] Performing attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding implicit representation vectors of the context semantic association of evaluation data at the first time scale and implicit representation vectors of the context semantic association of evaluation data at the second time scale in the time queue of the context semantic association of evaluation data at the first time scale and implicit representation vectors of the context semantic association of evaluation data at the second time scale, so as to obtain a time queue of implicit representation vectors of multi-scale temporal context semantic association of evaluation data;

[0011] Based on the time queue of the multi-scale temporal contextual semantic association implicit representation vector of the evaluation data, the customer's emotion polarity recognition result is determined.

[0012] Specifically, a multi-scale temporal context analysis is performed on the set of evaluation data of the franchise stores to obtain a time queue of a first time scale evaluation data context semantic association implicit representation vector and a time queue of a second time scale evaluation data context semantic association implicit representation vector, including:

[0013] Arranging each evaluation data in the set of evaluation data according to the timestamp to obtain a time queue of the evaluation data;

[0014] Performing semantic embedding coding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding coding vectors of the evaluation data;

[0015] The time queue of the evaluation data semantic embedding encoding vector is input into the evaluation data sequence encoder based on the RNN-LSTM hybrid model to obtain the time queue of the evaluation data context semantic association implicit representation vector of the first time scale and the time queue of the evaluation data context semantic association implicit representation vector of the second time scale.

[0016] Specifically, performing semantic embedding coding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding coding vectors of the evaluation data includes:

[0017] A semantic embedding encoder based on a Transformer architecture is used to perform semantic embedding encoding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding encoding vectors of the evaluation data.

[0018] Specifically, the time queue of the first time scale evaluation data context semantic association implicit representation vector and the time queue of the second time scale evaluation data context semantic association implicit representation vector corresponding to each group in the time queue of the first time scale evaluation data context semantic association implicit representation vector and the second time scale evaluation data context semantic association implicit representation vector are respectively subjected to attention interaction fusion guided by implicit collaborative features to obtain the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector, including:

[0019] Inputting the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale corresponding to the timestamp into the collaborative feature extraction network to obtain an implicit collaborative encoding vector between multi-scale temporal features of the evaluation data;

[0020] Based on the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data, feature attention modulation is performed on the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale to obtain an optimized implicit representation vector of the context semantic association of the evaluation data of the first time scale and an optimized implicit representation vector of the context semantic association of the evaluation data of the second time scale;

[0021] The optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector are fused to obtain the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector.

[0022] Specifically, the collaborative feature extraction network includes three parallel feature interaction layers, a multi-level interactive feature cascade layer, a point convolution layer and an activation layer based on the Leaky ReLU function.

[0023] Specifically, based on the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data, feature attention modulation is performed on the implicit representation vector of the context semantic association of the first time scale evaluation data and the implicit representation vector of the context semantic association of the second time scale evaluation data to obtain the optimized implicit representation vector of the context semantic association of the first time scale evaluation data and the optimized implicit representation vector of the context semantic association of the second time scale evaluation data, including:

[0024] Writing the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data into a dynamic memory unit to obtain a dynamic key vector;

[0025] Extract the dynamic key vector from the dynamic memory unit, and input the first time scale evaluation data context semantic association implicit representation vector and the dynamic key vector into a feature attention modulation module based on a first converter structure to obtain the optimized first time scale evaluation data context semantic association implicit representation vector;

[0026] The dynamic key vector is extracted from the dynamic memory unit, and the second time scale evaluation data context semantic association implicit representation vector and the dynamic key vector are input into a feature attention modulation module based on a second converter structure to obtain the optimized second time scale evaluation data context semantic association implicit representation vector.

[0027] Specifically, fusing the optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector to obtain the evaluation data multi-scale temporal context semantic association implicit representation vector, including:

[0028] The optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector are input into a feature linear interaction network for weighted fusion to obtain the evaluation data multi-scale temporal context semantic association implicit representation vector.

[0029] Specifically, based on the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data, determining the customer's emotion polarity recognition result includes:

[0030] The time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data is input into the classifier-based customer emotion polarity identifier to obtain the emotion polarity identification result, and the emotion polarity identification result is used to indicate whether the customer's emotion polarity towards the franchise store is changing positively or negatively.

[0031] According to another aspect of the present application, a franchise store operation optimization system based on big data is provided, which includes:

[0032] An evaluation data extraction module is used to extract a set of evaluation data for franchise stores from customer feedback data;

[0033] A multi-scale time series analysis module, used for performing a multi-scale time series context analysis on the set of evaluation data of the franchise stores to obtain a time queue of a first time scale evaluation data context semantic association implicit representation vector and a time queue of a second time scale evaluation data context semantic association implicit representation vector;

[0034] An attention interaction fusion module is used to perform attention interaction fusion guided by implicit collaborative features on each group of corresponding first time scale evaluation data context semantic association implicit representation vectors and second time scale evaluation data context semantic association implicit representation vectors in the time queue of the first time scale evaluation data context semantic association implicit representation vectors and the time queue of the second time scale evaluation data context semantic association implicit representation vectors, respectively, to obtain a time queue of evaluation data multi-scale temporal context semantic association implicit representation vectors;

[0035] The emotion polarity recognition module is used to determine the emotion polarity recognition result of the customer based on the time queue of the implicit representation vector of the multi-scale temporal context semantic association of the evaluation data.

[0036] This application has at least the following technical effects:

[0037] Compared with the prior art, the franchise store operation optimization method and system based on big data provided by the present application first extracts a set of evaluation data for franchise stores from customer feedback data, and uses data processing technology based on deep learning to semantically encode each evaluation data, extracts the semantic embedding feature representation of each evaluation data, and then, through multi-scale temporal context semantic association analysis of each evaluation data, to dig out the implicit customer emotion change trend in the evaluation data, thereby realizing intelligent recognition of customer emotion polarity changes. In this way, customer needs and preferences can be more accurately grasped, thereby adjusting menus, promotion strategies and business strategies to improve customer satisfaction and loyalty, and thus achieve better business benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0039] Figure 1 This is a flow chart of a franchise store operation optimization method based on big data according to an embodiment of the present application.

[0040] Figure 2 This is a data flow diagram of a franchise store operation optimization method based on big data according to an embodiment of the present application.

[0041] Figure 3 This is a flowchart of sub-step S2 of the franchise store operation optimization method based on big data according to an embodiment of the present application.

[0042] Figure 4 This is a flowchart of sub-step S3 of the franchise store operation optimization method based on big data according to an embodiment of the present application.

[0043] Figure 5 This is a flowchart of sub-step S32 of the franchise store operation optimization method based on big data according to an embodiment of the present application.

[0044] Figure 6 It is a block diagram of a franchise store operation optimization system based on big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0046] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0047] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0048] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0049] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where the data is located, and with the authorization given by the owner of the corresponding device.

[0050] In response to the technical problems described in the above background technology, the present application proposes an optimized franchise store operation optimization method based on big data. The method first extracts a set of evaluation data for franchise stores from customer feedback data, and uses deep learning-based data processing technology to semantically encode each evaluation data, and extracts the semantic embedded feature representation of each evaluation data. Then, by performing multi-scale temporal context semantic association analysis on each evaluation data, the customer emotion change trend implicit in the evaluation data is mined, thereby realizing intelligent identification of customer emotion polarity changes, which can more accurately grasp customer needs and preferences, thereby adjusting menus, promotion strategies and business strategies to improve customer satisfaction and loyalty, and thus achieve better operating benefits.

[0051] Figure 1 This is a flow chart of a franchise store operation optimization method based on big data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the franchise store operation optimization method based on big data according to an embodiment of the present application. Figure 1 and Figure 2As shown, the franchise store operation optimization method based on big data includes the steps of: S1, extracting a set of evaluation data for franchise stores from customer feedback data; S2, performing multi-scale temporal context analysis on the set of evaluation data of franchise stores to obtain a time queue of first time scale evaluation data context semantic association implicit representation vectors and a time queue of second time scale evaluation data context semantic association implicit representation vectors; S3, performing attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding first time scale evaluation data context semantic association implicit representation vectors and second time scale evaluation data context semantic association implicit representation vectors in the time queue of first time scale evaluation data context semantic association implicit representation vectors and the time queue of second time scale evaluation data context semantic association implicit representation vectors to obtain a time queue of multi-scale temporal context semantic association implicit representation vectors of evaluation data; S4, determining the customer's emotion polarity recognition result based on the time queue of multi-scale temporal context semantic association implicit representation vectors of evaluation data.

[0052] In the above franchise store operation optimization method based on big data, the step S1 extracts a set of evaluation data for the franchise store from the customer feedback data, which may be evaluation data of dishes, or evaluation indicators of service attitude, environment atmosphere, etc.

[0053] In the above franchise store operation optimization method based on big data, obtaining customer feedback data is the first and most important step in the entire process. This process requires the use of scientific and reasonable methods to extract valuable evaluation information for specific dishes from massive amounts of data. The specific implementation method of this process will be described in detail below:

[0054] In order to more effectively analyze the emotional changes of customers towards franchise stores, it is first necessary to accurately extract all the evaluation data for the dish from the customer feedback data. With the popularization of the Internet and mobile applications, online evaluation has become an important way for consumers to share their dining experience. Therefore, it is a very direct and effective way to capture relevant comments from major catering evaluation websites (such as Dianping, Yelp, etc.), social media platforms (such as Weibo, WeChat Moments, Douyin, etc.) and the brand's own official website or APP. Web crawler technology plays an important role in this situation. It is an automated program that can automatically access the target website according to the set rules and extract the required information. When using web crawlers, it is necessary to comply with the terms of service of the target website, not to conduct any behavior that may violate its use policy, and to ensure that the data collection process complies with the requirements of local laws and regulations and respects user privacy. In addition, since the structure of the web page may change, the crawler script needs to be updated regularly to adapt to such changes to ensure continuous and stable acquisition of the latest data.

[0055] Secondly, offline questionnaires are also an important means of obtaining customer feedback. Although online reviews account for a large proportion, some customers are still accustomed to providing feedback by filling out paper questionnaires or participating in face-to-face interviews. To this end, a standardized survey form can be designed to cover evaluation indicators such as food quality, service attitude, and environmental atmosphere, and customers are encouraged to complete it immediately after dining. At the same time, a suggestion box can be set up for customers to submit suggestions or complaints anonymously. All collected handwritten materials must be scanned and converted into electronic document form for subsequent unified management and analysis. This not only retains the advantages of traditional methods, but also facilitates centralized data processing.

[0056] In addition, immediate feedback directly from customers cannot be ignored. Sometimes, customers will express their opinions through telephone, email or even face-to-face communication. To this end, franchise stores should establish a complete customer relationship management system (CRM) to record and track each customer's contact information and historical interaction records. At the same time, employees should be trained to master basic data entry skills to ensure that each customer's opinions can be recorded in a timely and accurate manner. This approach not only enhances customer participation, but also enables companies to respond to customer needs and concerns more quickly.

[0057] Furthermore, sometimes companies may choose to cooperate with professional market research companies and entrust them to conduct large-scale customer satisfaction surveys on a regular basis. This method can obtain more comprehensive and in-depth industry insight reports, although the cost is relatively high. In addition, joining industry associations or alliance organizations to share the experience and resource base accumulated among members is also an effective supplementary means.

[0058] Finally, many chain brands have established their own membership systems and accumulated a large amount of user behavior data. By analyzing these historical order records, points redemption information and other information, we can also indirectly understand customers' preferences for different dishes and their consumption preferences.

[0059] After data collection is completed, the collected customer feedback data needs to be cleaned and preprocessed to remove invalid or irrelevant information to ensure the accuracy of subsequent analysis. For example, some comments may contain content that is not related to the dishes, such as service quality, environmental atmosphere, etc. At this time, it is necessary to use natural language processing technology to identify and remove this part of information. In addition, it is also necessary to exclude duplicate submissions and obvious spam to avoid affecting the overall analysis results. In terms of specific operations, you can use the pandas library in Python to perform basic data preprocessing, such as deleting empty value rows, filtering specific keywords, etc.

[0060] Next comes the text classification phase, the purpose of which is to categorize all the reviews about the franchise stores individually. Keyword search can be used here. Specifically, first define clear screening criteria, including determining keywords or phrases related to the franchise stores, such as dish names, aliases, and descriptive words. Then, using these screening criteria, the collected review data is automatically classified through text analysis tools. In addition, in order to ensure the quality of the screening results, testing and verification are indispensable. This includes randomly selecting some of the screened data for manual review to ensure the accuracy of the classification.

[0061] In the above-mentioned franchise store operation optimization method based on big data, the step S2 performs multi-scale temporal context analysis on the set of evaluation data of the franchise stores to obtain a time queue of the context semantic association implicit representation vector of the first time scale evaluation data and a time queue of the context semantic association implicit representation vector of the second time scale evaluation data. Figure 3 FIG. 1 is a flowchart of sub-step S2 of the franchise store operation optimization method based on big data according to an embodiment of the present application. Figure 3 As shown, the step S2 includes the steps of: S21, arranging each evaluation data in the set of evaluation data according to timestamps to obtain a time queue of evaluation data; S22, performing semantic embedding encoding on each evaluation data in the time queue of evaluation data to obtain a time queue of semantic embedding encoding vectors of evaluation data; S23, inputting the time queue of semantic embedding encoding vectors of evaluation data into an evaluation data sequence encoder based on an RNN-LSTM hybrid model to obtain a time queue of contextual semantically associated implicit representation vectors of evaluation data of the first time scale and a time queue of contextual semantically associated implicit representation vectors of evaluation data of the second time scale.

[0062] Specifically, in step S21, each evaluation data in the set of evaluation data is arranged according to the timestamp to obtain a time queue of the evaluation data. Among them, the present application takes into account that the customer's emotions and feedback on the dishes will change over time. For example, changes in seasonal ingredients, changes in chefs, or the holding of special events may affect the customer's dining experience. Therefore, in order to maintain the temporal logical coherence of the data and dig out valuable information hidden in the time dimension, the present application further arranges each evaluation data in the set of evaluation data according to the timestamp to form a time queue of the evaluation data. In this way, it is helpful to understand the development trend of customer emotions over time and help establish causal relationships between events. For example, if a large number of negative reviews appear in a certain time period, managers can determine the problem by reviewing the events that occurred during that period (such as the launch of a new menu, price adjustments, etc.) and take corresponding corrective measures, thereby helping franchise stores better understand and adapt to changes in customer needs.

[0063] Specifically, in a specific example of the present application, the step S22 includes: using a semantic embedding encoder based on the Transformer architecture to perform semantic embedding encoding on each evaluation data in the time queue of the evaluation data to obtain a time queue of the semantic embedding encoding vector of the evaluation data. Considering that the traditional keyword-based sentiment classification method cannot deeply understand the complex context and emotions of customer feedback, it may cause inaccurate sentiment classification. Therefore, the present application further uses a semantic embedding encoder based on the Transformer architecture to perform semantic embedding encoding on each evaluation data in the time queue of the evaluation data, so as to utilize the self-attention mechanism of the Transformer architecture to capture the long-distance semantic dependency between each word in the evaluation data, and more accurately understand the contextual information in each evaluation data, such as the customer's specific description of the dish, emotional expression, etc., thereby generating a time queue of the semantic embedding encoding vector of the evaluation data.

[0064] In the above process of using the semantic embedding encoder based on the Transformer architecture to process the evaluation data to obtain the time queue of the semantic embedding encoding vector of the evaluation data, multiple links are involved, such as text preprocessing, model training, encoding, and time series organization. The specific implementation method of the whole process will be described in detail below:

[0065] First, you need to prepare the evaluation data set to be processed. This data may come from user comments on products or services, or posts on social media. Each evaluation data usually contains one or more paragraphs of text information and a timestamp associated with it. In order to ensure the effectiveness of subsequent processing, the original data must be cleaned and preprocessed, including removing irrelevant characters, standardizing text formats (such as unified capitalization), and word segmentation.

[0066] Next, select a suitable pre-trained model based on the Transformer architecture as the basis. After selecting the model, you can fine-tune it to adapt to the corpus characteristics of a specific field and improve the encoding quality. During the fine-tuning process, you can use a portion of the annotated data set as training samples and define an appropriate loss function to guide the learning process. It is worth noting that in actual applications, considering the limitations of computing resources, sometimes the basic model without fine-tuning is directly used for encoding.

[0067] Once the model is ready, you can start to perform semantic embedding encoding on each evaluation data. Specifically, for each input text, it is first converted into a series of tokens through a tokenizer, and then sent to the pre-trained Transformer model. The model generates a fixed-length vector representation for each token, reflecting its context-related semantic information. Finally, by performing some form of aggregation on all token-level vectors (such as averaging or maximum pooling), the contextual semantic feature representation of each evaluation data can be obtained, and the time queue of the semantic embedding encoding vector of the evaluation data can be obtained.

[0068] After obtaining the semantic embedding encoding vectors corresponding to a series of evaluation data, it is necessary to sort them according to the timestamp of each evaluation data. This step ensures that the final output vector sequence can reflect the information pattern that changes over time. In practice, the pandas library in Python can be used to implement the function of sorting data by time.

[0069] In summary, by using the semantic embedding encoder based on the Transformer architecture to process the time queue of evaluation data, we can not only obtain richer and more accurate information than traditional methods, but also better support various subsequent application requirements.

[0070] Specifically, in step S23, the time queue of the evaluation data semantic embedding encoding vector is input into the evaluation data sequence encoder based on the RNN-LSTM hybrid model to obtain the time queue of the first time scale evaluation data context semantic association implicit representation vector and the time queue of the second time scale evaluation data context semantic association implicit representation vector. Among them, the customer's emotional fluctuations in a short period of time to the dishes may be related to a specific event (such as a promotional activity), while the long-term emotional trend reflects a deeper level of brand loyalty or product quality stability. Therefore, in order to improve the analysis accuracy of customer emotional changes, the present application adopts a multi-scale temporal context semantic association analysis method, and processes the time queue of the evaluation data semantic embedding encoding vector by constructing an RNN-LSTM hybrid model to simultaneously capture the short-term and long-term trends of customer emotional changes. It should be known to those of ordinary skill in the art that the RNN model has excellent sequence data processing capabilities and can effectively use the context information in the sequence to optimize the feature representation of each time step, but it may encounter the problem of gradient disappearance or explosion when processing long sequence data, and the LSTM model, as a variant of the RNN model, effectively solves this problem by introducing a gating mechanism, and can effectively learn the long-term dependency information in the sequence data. Therefore, in the technical solution of the present application, the time queue of the semantic embedded coding vector of the evaluation data is respectively analyzed in time series by using the RNN model and the LSTM model to generate a time queue of the implicit representation vector of the context semantic association of the evaluation data at the first time scale and a time queue of the implicit representation vector of the context semantic association of the evaluation data at the second time scale, thereby being able to take into account the dual needs of capturing short-term emotional fluctuations and long-term emotional trends, and more comprehensively understand the dynamic changes of customer emotions.

[0071] Specifically, RNN is a neural network architecture that is particularly suitable for processing sequence data. Unlike traditional feedforward neural networks, RNN has an internal state or memory that allows it to retain some information about what it has seen before when processing each element in the sequence. This makes RNN very suitable for processing time series analysis. However, standard RNN has a major limitation - the gradient disappearance problem, that is, when trying to learn long-term dependencies, the gradient may become so small that the weights cannot be updated effectively, thus affecting model performance.

[0072] In order to solve the above-mentioned gradient vanishing problem, the LSTM structure is introduced. LSTM is a special form of RNN, which controls the information flow by introducing a gating mechanism, including components such as forget gate, input gate and output gate. These gates can help the model decide which information should be remembered, which information can be forgotten, and when to output the results. This design enables LSTM to ignore irrelevant information while keeping important historical information, thus effectively solving the gradient vanishing problem faced by traditional RNN, and performing better in handling dependencies over long time spans.

[0073] Therefore, in order to better capture the emotional change patterns contained in customer evaluation data, this application combines the advantages of RNN and LSTM to build a hybrid model, first using a simple RNN layer to capture dynamic changes on a shorter time scale, and then further extracting deeper long-term trends through subsequent LSTM layers. In this way, the RNN-LSTM hybrid model can effectively understand the dynamic change patterns of customer emotions on different time scales.

[0074] In the above-mentioned franchise store operation optimization method based on big data, the step S3 is to perform attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding first time scale evaluation data context semantic association implicit representation vectors and second time scale evaluation data context semantic association implicit representation vectors in the time queue of the first time scale evaluation data context semantic association implicit representation vectors and the time queue of the second time scale evaluation data context semantic association implicit representation vectors to obtain a time queue of evaluation data multi-scale temporal context semantic association implicit representation vectors. Among them, considering that the emotion change patterns of different time scales provide complementary information, in order to more comprehensively understand the emotional changes of customers, the present application further fuses the time queue of the first time scale evaluation data context semantic association implicit representation vectors and the time queue of the second time scale evaluation data context semantic association implicit representation vectors according to the timestamp, so as to comprehensively consider the short-term fluctuations and long-term trends of customer emotions, and generate a time queue of evaluation data multi-scale temporal context semantic association implicit representation vectors. In particular, in order to optimize the feature fusion effect, this application proposes a fusion strategy based on the attention mechanism, which guides and strengthens the key information in the implicit representation vectors of the context semantic associations of evaluation data at different time scales by mining the implicit collaborative features between the implicit representation vectors of the context semantic associations of evaluation data at different time scales, thereby generating a more accurate comprehensive feature representation of customer sentiment changes and improving the accuracy of sentiment analysis.

[0075] Figure 4 FIG. 1 is a flowchart of sub-step S3 of the franchise store operation optimization method based on big data according to an embodiment of the present application. Figure 4As shown, the step S3 includes the steps of: S31, inputting the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale corresponding to the timestamp into the collaborative feature extraction network to obtain the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data; S32, based on the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data, performing feature attention modulation on the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale respectively to obtain the optimized implicit representation vector of the context semantic association of the evaluation data of the first time scale and the optimized implicit representation vector of the context semantic association of the evaluation data of the second time scale; S33, fusing the optimized implicit representation vector of the context semantic association of the evaluation data of the first time scale and the optimized implicit representation vector of the context semantic association of the evaluation data of the second time scale to obtain the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data.

[0076] Specifically, the step S31 inputs the implicit representation vector of the context semantic association of the first time scale evaluation data and the implicit representation vector of the context semantic association of the second time scale evaluation data corresponding to the timestamp into the collaborative feature extraction network to obtain an implicit collaborative coding vector between multi-scale temporal features of the evaluation data. In a specific example of the present application, the collaborative feature extraction network includes three parallel feature interaction layers, a multi-level interactive feature cascade layer, a point convolution layer, and an activation layer based on the Leaky ReLU function, which is expressed by the formula:

[0077]

[0078] Wherein, V1 represents the implicit representation vector of the context semantic association of the evaluation data of the first time scale, V2 represents the implicit representation vector of the context semantic association of the evaluation data of the second time scale, ⊕ represents point addition by position, ⊙ represents point multiplication by position, represents point-by-point subtraction, concat(·;·;·) represents feature concatenation, Conv 1×1 represents a 1×1 convolution operation, Leaky ReLU represents the Leaky ReLU activation function, and V J Represents the implicit collaborative coding vector between multi-scale temporal features of evaluation data.

[0079] That is, the implicit representation vector of the context semantic association of the first time scale evaluation data and the implicit representation vector of the context semantic association of the second time scale evaluation data are input into the collaborative feature extraction network, and through multi-level interaction, point convolution and non-linear activation operations, the deep correlation between the features is captured to mine the implicit collaborative feature representation between the two.

[0080] Figure 5 FIG. 1 is a flowchart of sub-step S32 of the franchise store operation optimization method based on big data according to an embodiment of the present application. Figure 5 As shown, the step S32 includes the steps of: S321, writing the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data into a dynamic memory unit to obtain a dynamic key vector; S322, extracting the dynamic key vector from the dynamic memory unit, and inputting the implicit representation vector of the context semantic association of the first time scale evaluation data and the dynamic key vector into a feature attention modulation module based on a first converter structure to obtain the optimized implicit representation vector of the context semantic association of the first time scale evaluation data; S323, extracting the dynamic key vector from the dynamic memory unit, and inputting the implicit representation vector of the context semantic association of the second time scale evaluation data and the dynamic key vector into a feature attention modulation module based on a second converter structure to obtain the optimized implicit representation vector of the context semantic association of the second time scale evaluation data.

[0081] More specifically, the step S321 is expressed by the formula:

[0082] m u =M(V J )

[0083] Where M(·) represents a 1×1 convolution operation, m u Represents a dynamic key vector.

[0084] Among them, the dynamic memory unit is a special type of memory structure, which works in a similar way to the memory unit in a recursive neural network. It can dynamically adjust its internal state according to the current input features to store and update long-term information related to changes in customer emotions, thereby ensuring that the model can take into account the previously accumulated emotional trends and patterns when processing the next set of contextual semantic association implicit representation vectors of the first time scale and the second time scale of the evaluation data, so as to better adapt to the contextual needs of the current scene.

[0085] More specifically, the step S322 is expressed by the formula:

[0086]

[0087] Among them, W 1$ and W 1v denote the first query embedding matrix and the first value embedding matrix respectively, V 1$ and V 1v denote the first query vector and the first value vector, respectively, and b 1$ and b 1vThey represent different bias items respectively, softmax(·) represents the normalized exponential function, S represents the characteristic scale value of the dynamic key vector, and V′1 represents the implicit representation vector of the context semantic association of the evaluation data at the first time scale after optimization.

[0088] More specifically, the step S323 is expressed by the formula:

[0089]

[0090] Among them, W 2$ and W 2v denote the second query embedding matrix and the second value embedding matrix, V 2$ and V 2v denote the second query vector and the second value vector, respectively, and b 2$ and b 2v Represent different bias terms, V2 ′ Represents the implicit representation vector of the semantic association of the contextual evaluation data of the second time scale after optimization.

[0091] Specifically, a query vector and a value vector for the implicit representation vector of the context semantic association of the evaluation data at the first time scale, and a query vector and a value vector for the implicit representation vector of the context semantic association of the evaluation data at the second time scale are respectively constructed, and the self-attention mechanism in the converter structure is used to perform query attention-based feature modulation on the implicit representation vector of the context semantic association of the evaluation data at the first time scale and the implicit representation vector of the context semantic association of the evaluation data at the second time scale based on the dynamic key vector, which is equivalent to using the dynamic key vector as a common attention space to enhance the expression of key information in the original features and reduce the influence of noise and irrelevant information. At the same time, it also makes the optimized implicit representation vector of the context semantic association of the evaluation data at the first time scale and the implicit representation vector of the context semantic association of the evaluation data at the second time scale be in an aligned semantic space to facilitate the subsequent feature interaction and fusion processing between the two.

[0092] Specifically, the step S33 includes: inputting the optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector into a feature linear interaction network for weighted fusion to obtain the evaluation data multi-scale temporal context semantic association implicit representation vector, which is expressed as:

[0093] V6=αV′1+(1-α)V′2

[0094] Among them, α is the fusion weight parameter, and V6 represents the implicit representation vector of the multi-scale temporal context semantic association of the evaluation data.

[0095] That is, a feature linear interaction network is used to perform weighted fusion on the optimized implicit representation vector of the context semantic association of the evaluation data of the first time scale and the optimized implicit representation vector of the context semantic association of the evaluation data of the second time scale, and further feature integration is performed through linear transformation to generate a multi-scale temporal context semantic association implicit representation vector of the evaluation data.

[0096] By performing the above processing on the contextual semantic association implicit representation vectors of the evaluation data of the first time scale and the second time scale corresponding to each time stamp, a time queue of multi-scale temporal contextual semantic association implicit representation vectors of the evaluation data that integrates the short-term fluctuations and long-term trend information of customers' emotions can be obtained, thereby providing richer and more accurate feature basis for subsequent customer sentiment polarity analysis.

[0097] In the above-mentioned franchise store operation optimization method based on big data, the step S4 determines the customer's emotion polarity recognition result based on the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data. In a specific example of the present application, the step S4 includes: inputting the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data into a customer emotion polarity identifier based on a classifier to obtain the emotion polarity recognition result, and the emotion polarity recognition result is used to indicate whether the customer's emotion polarity for the franchise store is positively or negatively transformed. Specifically, first, the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data is arranged into a one-dimensional vector representation according to the time dimension to maintain the continuity and time sequence of the customer's emotion change, and then it is input as an input feature to a pre-trained classifier for emotion polarity analysis. The classifier learns the changing trend of the customer's emotion-related features in the time dimension by performing multi-layer neural network processing on the input features to identify the positive or negative change of the customer's emotion polarity. At the output of the classifier, a softmax function is set to calculate the probability distribution of the customer's emotional polarity change, so as to reflect the possibility of the customer's emotional polarity change. By comparing the probability values ​​of different emotional polarity categories, the final recognition result of the customer's emotional polarity can be determined. For example, if the probability of positive emotional polarity change is higher than the probability of negative emotional polarity change, it can be judged that the customer's emotion towards the dish is positive. On the contrary, it is a negative change. In this way, franchise stores can accurately understand the emotional changes of customers towards dishes and then adopt corresponding operation strategies. For example, if the recognition result shows that the emotional polarity of customers towards a certain dish is generally negative, then the preparation process or ingredient selection of the dish may need to be improved. In this way, franchise stores can better meet customer needs, improve customer satisfaction, and thus gain an advantage in the fiercely competitive market.

[0098] In particular, when each evaluation data multi-scale temporal context semantic association implicit representation vector in the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector respectively represents the multi-scale temporal semantic association feature of the evaluation data in the local time domain window, when the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector is input into the customer emotion polarity identifier based on the classifier, the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector is first spliced ​​into a full-time domain evaluation data semantic association feature vector, and then input into the customer emotion polarity identifier based on the classifier to obtain the emotion polarity recognition result. However, considering that there are encoding differences in the emotion polarity temporal expressions of the evaluation data in each local time domain, simple feature splicing will cause the overall feature manifold of the full-time domain evaluation data semantic association feature vector to have splicing folds in the high-dimensional feature space, that is, the overall feature manifold of the full-time domain evaluation data semantic association feature vector is not sufficiently smooth in the high-dimensional space, which affects its accuracy in classification probability regression.

[0099] Based on this, when the semantic association feature vector of the full-time domain evaluation data is input into the customer emotion polarity identifier based on the classifier for classification probability regression, the application first optimizes the semantic association feature vector of the full-time domain evaluation data, and the optimization process specifically includes:

[0100] Determine the mean and variance of the feature set composed of the feature values ​​of the semantic association feature vector of the full-time domain evaluation data, and divide the mean by the variance to obtain the semantic association probability weight value of the full-time domain evaluation data, which is expressed as follows:

[0101]

[0102] Among them, θ is the probability weight value of the semantic association of the full-time domain evaluation data, and μ and σ respectively represent the mean and variance of the feature set composed of the eigenvalues ​​of the semantic association feature vector of the full-time domain evaluation data.

[0103] The semantic association feature vector of the full-time domain evaluation data is multiplied by the inverse of the maximum eigenvalue of the semantic association feature vector of the full-time domain evaluation data to obtain the semantic association probability constraint vector of the full-time domain evaluation data, which is expressed as follows:

[0104] F1=F⊙f >a@ A1

[0105] Among them, F1 is the semantic association probability constraint vector of the full-time domain evaluation data, F is the semantic association feature vector of the full-time domain evaluation data, ⊙ represents the point multiplication by position, and f >a@ A1It is the inverse of the maximum eigenvalue of the semantic association feature vector of the full-time domain evaluation data.

[0106] The semantic association probability constraint vector of the full-time domain evaluation data is added with the semantic association probability weight value of the full-time domain evaluation data, and the logarithmic value with base 2 is calculated to obtain the semantic association information interaction vector of the full-time domain evaluation data, which is expressed as follows:

[0107]

[0108] Among them, F2 is the semantic association information interaction vector of the full-time domain evaluation data, represents point-wise addition, and log2(·) represents the logarithmic function with base 2.

[0109] After subtracting the full-time-domain evaluation data semantic association probability constraint vector from the full-time-domain evaluation data semantic association probability weight value, the inverse of each eigenvalue is calculated to obtain the full-time-domain evaluation data semantic association sequence constraint vector, which is expressed as follows:

[0110]

[0111] in,· D is the semantic association sequence constraint vector of the full-time evaluation data, Indicates point-by-point subtraction, (·) A1 Represents the reciprocal of the eigenvalue.

[0112] Finally, the full-time-domain evaluation data semantic association information interaction vector is added to the full-time-domain evaluation data semantic association sequence constraint vector to obtain an optimized full-time-domain evaluation data semantic association feature vector.

[0113] This application takes into account that when the weight matrix during classification probability regression acts on the semantic association feature vector of the full-time domain evaluation data, the distribution diversity of the semantic association feature vector of the full-time domain evaluation data causes the uncertainty of the weight stimulus parameters of the weight matrix, which makes the posterior probability density inference of the semantic association feature vector of the full-time domain evaluation data through the weight matrix missing, which will affect the accuracy of the probability regression result. The above-mentioned optimization process simulates the mesoscale interaction structure under probability constraints between the eigenvalue scale and the eigenvector scale of the semantic association feature vector of the full-time domain evaluation data through the probability statistical characteristics of the semantic association feature vector of the full-time domain evaluation data, so as to construct a bidirectional hidden variable motif based on the mesoscale short sequence relative to the probability statistical value to perform mesoscale bidirectional migration, and perform posterior recovery based on the sequence constraints on the interactive information, so as to improve the convergence effect of the semantic association feature vector of the full-time domain evaluation data in the probability density domain, and improve the accuracy of the emotion polarity recognition result obtained by inputting the customer emotion polarity identifier based on the classifier.

[0114] In summary, the big data-based franchise store operation optimization method based on the embodiment of the present application is explained, which first extracts a set of evaluation data for franchise stores from customer feedback data, and uses data processing technology based on deep learning to semantically encode each evaluation data, extracts the semantic embedding feature representation of each evaluation data, and then, through multi-scale temporal context semantic association analysis of each evaluation data, to dig out the implicit customer emotion change trend in the evaluation data, so as to realize intelligent recognition of customer emotion polarity change. In this way, the needs and preferences of customers can be grasped more accurately, so as to adjust the menu, promotion strategy and business strategy to improve customer satisfaction and loyalty, and then achieve better business benefits.

[0115] Furthermore, a franchise store operation optimization system based on big data is also provided.

[0116] Figure 6 FIG. 1 is a block diagram of a franchise store operation optimization system based on big data according to an embodiment of the present application. Figure 6 As shown, according to the embodiment of the present application, the franchise store operation optimization system 100 based on big data includes: an evaluation data extraction module 110, which is used to extract a set of evaluation data for franchise stores from customer feedback data; a multi-scale time series analysis module 120, which is used to perform multi-scale time series context analysis on the set of evaluation data of the franchise stores to obtain a time queue of first time scale evaluation data context semantic association implicit representation vectors and a time queue of second time scale evaluation data context semantic association implicit representation vectors; an attention interaction fusion module 130, which is used to perform attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding first time scale evaluation data context semantic association implicit representation vectors and second time scale evaluation data context semantic association implicit representation vectors in the time queue of the first time scale evaluation data context semantic association implicit representation vectors and the time queue of the second time scale evaluation data context semantic association implicit representation vectors to obtain a time queue of multi-scale time series context semantic association implicit representation vectors of evaluation data; an emotion polarity recognition module 140, which is used to determine the customer's emotion polarity recognition result based on the time queue of the multi-scale time series context semantic association implicit representation vectors of the evaluation data.

[0117] The specific operations of each module in the above franchise store operation optimization system based on big data have been referenced above. Figures 1 to 5 The description of the franchise store operation optimization method based on big data has been introduced in detail, and therefore, its repeated description will be omitted.

[0118] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0119] In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0120] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0121] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0122] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A franchise store operation optimization method based on big data, characterized in that: include: Extracting a collection of evaluation data for franchise stores from customer feedback data; Performing a multi-scale temporal context analysis on the set of evaluation data of the franchise stores to obtain a time sequence of a first time-scale evaluation data context semantic association implicit representation vector and a time sequence of a second time-scale evaluation data context semantic association implicit representation vector; Performing attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding implicit representation vectors of the context semantic association of evaluation data at the first time scale and implicit representation vectors of the context semantic association of evaluation data at the second time scale in the time queue of the context semantic association of evaluation data at the first time scale and implicit representation vectors of the context semantic association of evaluation data at the second time scale, so as to obtain a time queue of implicit representation vectors of multi-scale temporal context semantic association of evaluation data; Based on the time queue of the multi-scale temporal contextual semantic association implicit representation vector of the evaluation data, the customer's emotion polarity recognition result is determined.

2. The franchise store operation optimization method based on big data according to claim 1 is characterized in that: A multi-scale temporal context analysis is performed on the set of evaluation data of the franchise stores to obtain a time queue of a first time scale evaluation data context semantic association implicit representation vector and a time queue of a second time scale evaluation data context semantic association implicit representation vector, including: Arranging each evaluation data in the set of evaluation data according to the timestamp to obtain a time queue of the evaluation data; Performing semantic embedding coding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding coding vectors of the evaluation data; The time queue of the evaluation data semantic embedding encoding vector is input into the evaluation data sequence encoder based on the RNN-LSTM hybrid model to obtain the time queue of the evaluation data context semantic association implicit representation vector of the first time scale and the time queue of the evaluation data context semantic association implicit representation vector of the second time scale.

3. The franchise store operation optimization method based on big data according to claim 2 is characterized in that: Performing semantic embedding coding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding coding vectors of the evaluation data, including: A semantic embedding encoder based on a Transformer architecture is used to perform semantic embedding encoding on each evaluation data in the time queue of the evaluation data to obtain a time queue of semantic embedding encoding vectors of the evaluation data.

4. The franchise store operation optimization method based on big data according to claim 3 is characterized in that: The method comprises: performing attention interaction fusion based on implicit collaborative feature guidance on each group of the first time scale evaluation data context semantic association implicit representation vector and the second time scale evaluation data context semantic association implicit representation vector in the time queue of the first time scale evaluation data context semantic association implicit representation vector and the time queue of the second time scale evaluation data context semantic association implicit representation vector to obtain a time queue of evaluation data multi-scale temporal context semantic association implicit representation vector, including: Inputting the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale corresponding to the timestamp into the collaborative feature extraction network to obtain an implicit collaborative encoding vector between multi-scale temporal features of the evaluation data; Based on the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data, feature attention modulation is performed on the implicit representation vector of the context semantic association of the evaluation data of the first time scale and the implicit representation vector of the context semantic association of the evaluation data of the second time scale to obtain an optimized implicit representation vector of the context semantic association of the evaluation data of the first time scale and an optimized implicit representation vector of the context semantic association of the evaluation data of the second time scale; The optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector are fused to obtain the time queue of the evaluation data multi-scale temporal context semantic association implicit representation vector.

5. The franchise store operation optimization method based on big data according to claim 4 is characterized in that: The collaborative feature extraction network includes three parallel feature interaction layers, a multi-level interactive feature cascade layer, a point convolution layer and an activation layer based on the LeakyReLU function.

6. The franchise store operation optimization method based on big data according to claim 5 is characterized in that: Based on the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data, feature attention modulation is performed on the first time scale evaluation data context semantic association implicit representation vector and the second time scale evaluation data context semantic association implicit representation vector to obtain an optimized first time scale evaluation data context semantic association implicit representation vector and an optimized second time scale evaluation data context semantic association implicit representation vector, including: Writing the implicit collaborative coding vector between the multi-scale temporal features of the evaluation data into a dynamic memory unit to obtain a dynamic key vector; Extract the dynamic key vector from the dynamic memory unit, and input the first time scale evaluation data context semantic association implicit representation vector and the dynamic key vector into a feature attention modulation module based on a first converter structure to obtain the optimized first time scale evaluation data context semantic association implicit representation vector; The dynamic key vector is extracted from the dynamic memory unit, and the second time scale evaluation data context semantic association implicit representation vector and the dynamic key vector are input into a feature attention modulation module based on a second converter structure to obtain the optimized second time scale evaluation data context semantic association implicit representation vector.

7. The franchise store operation optimization method based on big data according to claim 6 is characterized in that: The optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector are integrated to obtain the evaluation data multi-scale temporal context semantic association implicit representation vector, including: The optimized first time scale evaluation data context semantic association implicit representation vector and the optimized second time scale evaluation data context semantic association implicit representation vector are input into a feature linear interaction network for weighted fusion to obtain the evaluation data multi-scale temporal context semantic association implicit representation vector.

8. The franchise store operation optimization method based on big data according to claim 7 is characterized in that: Determining the customer's emotion polarity recognition result based on the time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data includes: The time queue of the multi-scale temporal context semantic association implicit representation vector of the evaluation data is input into the classifier-based customer emotion polarity identifier to obtain the emotion polarity identification result, and the emotion polarity identification result is used to indicate whether the customer's emotion polarity towards the franchise store is changing positively or negatively.

9. The franchise store operation optimization system based on big data is characterized by: include: An evaluation data extraction module is used to extract a set of evaluation data for franchise stores from customer feedback data; A multi-scale time series analysis module, used for performing a multi-scale time series context analysis on the set of evaluation data of the franchise stores to obtain a time queue of a first time scale evaluation data context semantic association implicit representation vector and a time queue of a second time scale evaluation data context semantic association implicit representation vector; An attention interaction fusion module is used to perform attention interaction fusion based on implicit collaborative feature guidance on each group of corresponding first time scale evaluation data context semantic association implicit representation vectors and second time scale evaluation data context semantic association implicit representation vectors in the time queue of the first time scale evaluation data context semantic association implicit representation vectors and the time queue of the second time scale evaluation data context semantic association implicit representation vectors, so as to obtain a time queue of evaluation data multi-scale temporal context semantic association implicit representation vectors.

10. The franchise store operation optimization system based on big data according to claim 9 is characterized in that: Also includes: The emotion polarity recognition module is used to determine the emotion polarity recognition result of the customer based on the time queue of the implicit representation vector of the multi-scale temporal context semantic association of the evaluation data.

Images