Rectification strategy generation method and device based on self-inspection content, equipment and medium

Abstract

The invention discloses a rectification strategy generation method and device based on self-examination content, computer equipment and a computer readable storage medium, and the method comprises the steps: firstly screening out a personalized self-examination problem related to a self-examination region image through a rectification specification knowledge graph, taking the personalized self-examination problem as a candidate problem, improving the adaptation degree of the self-examination problem and a target place, and improving the self-examination efficiency. And then based on an entropy reduction principle, determining a target problem most related to the target place. The method comprises the following steps: acquiring a target question, performing interactive question answering with a user based on the target question, and proposing a targeted question to the user, so as to deeply mine dynamic description information of a self-examination result of the user on the target place as answer information, and finally inputting the dynamic description information into a strategy generation model to obtain a target rectification strategy corresponding to the target place. Therefore, the labor cost is reduced, the self-inspection efficiency and the generation efficiency of the rectification strategy are improved, and the self-inspection accuracy and the adaptation degree of the rectification strategy are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, computer equipment, and computer-readable storage medium for generating a rectification strategy based on self-inspection content. Background Art

[0002] At present, the food business information of each catering service unit is mainly checked manually, and a large number of catering service units rely entirely on manual inspection, which is inefficient. Therefore, how to improve the self-inspection efficiency of catering service units is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present application provides a method, apparatus, computer equipment, and computer-readable storage medium for generating a rectification strategy based on self-inspection content, aiming to improve the self-inspection efficiency and rectification strategy generation efficiency of catering service units.

[0004] In a first aspect, the present application provides a method for generating a rectification strategy based on self-examination content, the method comprising: Based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, the candidate questions corresponding to the target place are obtained; Based on the entropy reduction principle, determining a target question corresponding to the target place from the candidate questions, and performing an interactive self-examination inquiry to the user based on the target question to obtain answer information of the user; Based on the preset strategy generation model, basic business information and the answer information, a target rectification strategy for the target place is generated so that the user can rectify the target place based on the target rectification strategy.

[0005] The above-mentioned rectification strategy generation method based on self-inspection content firstly selects personalized self-inspection questions related to the self-inspection area image as candidate questions through the rectification specification knowledge map, thereby improving the adaptability of the self-inspection questions to the target place, and then determines the target question most relevant to the target place based on the entropy reduction principle. Then, based on the target question, interactive questions and answers are conducted with the user, and targeted questions are asked to the user, so as to deeply mine the dynamic description information of the user's self-inspection results of the target place as answer information, and finally the dynamic description information is input into the strategy generation model to obtain the target rectification strategy corresponding to the target place. As a result, not only the labor cost is reduced, the self-inspection efficiency and the rectification strategy generation efficiency are improved, but also the self-inspection accuracy and the adaptability of the rectification strategy are improved.

[0006] In a second aspect, the present application further provides a device for generating a rectification strategy based on self-examination content, the device comprising: A candidate question generation module, used to obtain candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user; A question interactive answering module, configured to determine a target question corresponding to the target place from the candidate questions based on the entropy reduction principle, and perform an interactive self-examination inquiry to the user based on the target question to obtain the user's answer information; The rectification strategy generation module is used to generate a target rectification strategy for the target place based on a preset strategy generation model, basic business information and the answer information, so that the user can rectify the target place based on the target rectification strategy.

[0007] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the above-mentioned method for generating a rectification strategy based on self-inspection content when executing the computer program.

[0008] In a fourth aspect, the present application further 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 processor implements the above-mentioned method for generating a rectification strategy based on self-inspection content. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0010] Figure 1 is a structural schematic diagram of a computer device provided in an embodiment of the present application; Figure 2 It is a schematic flow chart of a method for generating a rectification strategy based on self-examination content provided in an embodiment of the present application; Figure 3 It is an interface diagram of different issues corresponding to a self-check progress and a dedicated verification section provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0012] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0013] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

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

[0015] See also Figure 1 , Figure 1 1 is a schematic diagram of a computer device 100 provided in an embodiment of the present application. The computer device 100 may include a processor 1001 and a memory 1002, wherein the processor 1001 and the memory 1002 may be connected via a bus, and the bus may be any applicable bus such as an Inter-integrated Circuit (I2C) bus.

[0016] Exemplarily, the computer device 100 may be a server or a terminal. The server may 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 (CDN), and big data and artificial intelligence platforms. The terminal may be an electronic device such as a smartphone, a tablet computer, a laptop computer, and a desktop computer.

[0017] The memory 1002 may include a storage medium and an internal memory. The storage medium may store an operating system and a computer program. The computer program includes program instructions, which, when executed, may enable the processor 1001 to execute the method for generating a rectification strategy based on self-examination content described in any embodiment.

[0018] The processor 1001 is used to provide computing and control capabilities and support the operation of the entire computer device 100 .

[0019] The processor 1001 may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other types of processors such as programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components. A general-purpose processor may be a microprocessor, or a general-purpose processor may be any conventional processor.

[0020] In one embodiment, the processor 1001 is used to run a computer program stored in the memory 1002 to implement the following steps: Based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, the candidate questions corresponding to the target place are obtained; Based on the entropy reduction principle, determining a target question corresponding to the target place from the candidate questions, and performing an interactive self-examination inquiry to the user based on the target question to obtain answer information of the user; Based on the preset strategy generation model, basic business information and the answer information, a target rectification strategy for the target place is generated so that the user can rectify the target place based on the target rectification strategy.

[0021] In one embodiment, the processor 1001 is further configured to implement the following aspects in terms of generating the target rectification strategy for the target site based on the preset strategy generation model, the basic business information and the answer information: Initialize the model parameters of each network layer in the strategy generation model and generate the probability parameters of each network layer; Based on the probability parameters of each network layer, determine the network layer to be trained and the network layer to be optimized in each network layer; Based on the preset training sample set and the initial model parameters, the network layer to be optimized is skipped, and the network layer to be trained is trained to obtain the strategy generation model.

[0022] In one embodiment, the processor 1001 is further used to implement the method of determining the network layer to be trained and the network layer to be optimized in each network layer based on the probability parameters of each network layer: Obtain any value within a preset value range as a random parameter of a network layer; Compare the probability parameters of each network layer with its corresponding random parameters; Based on the comparison results of each network layer, the network layer whose probability parameter is not less than the random parameter is determined as the network layer to be trained, and the network layer whose probability parameter is less than the random parameter is determined as the network layer to be optimized.

[0023] In one embodiment, the processor 1001 is further used to implement the self-inspection area image of the target place uploaded by the user based on the rectification specification knowledge graph and the acquisition of candidate questions corresponding to the target place: Collect project information corresponding to various regulations and rules of various places, wherein the project information includes non-standard examples, standard examples, and rectification information from non-standard to standard; Segment the project information into at least one text unit, and extract at least one entity information and relationship information corresponding to each entity from each text unit based on a language model; Based on the entity information and the relationship information corresponding to each entity, the rectification specification knowledge graph is generated.

[0024] In one embodiment, the processor 1001, in realizing the generation of the rectification specification knowledge graph based on each of the entity information and the relationship information corresponding to each of the entities, is further configured to realize: Based on each of the entity information and the relationship information corresponding to each entity, construct a weighted undirected graph corresponding to the project information, wherein the nodes of the weighted undirected graph represent the entity information, and the edges of the weighted undirected graph represent the relationship information; The weighted undirected graph is hierarchically clustered to obtain information of each community, and based on the summary corresponding to each community, the rectification specification knowledge graph is generated.

[0025] In one embodiment, the processor 1001 is further used to implement the self-inspection area image of the target place uploaded by the user based on the rectification specification knowledge graph and the acquisition of candidate questions corresponding to the target place: Performing semantic analysis on the self-inspection area image to obtain identifications of each self-inspection area; Based on the self-inspection area identifier, determine the relevant regulations and their corresponding rectification graphic subsets in the rectification regulation knowledge graph; Based on the relevant regulations and their corresponding rectification graphic subsets, the self-inspection area image is analyzed to generate the candidate questions.

[0026] In one embodiment, the processor 1001 is further configured to implement the following aspects in terms of generating the target rectification strategy for the target site based on the preset strategy generation model, the basic business information and the answer information: When the target rectification strategy is to add necessary equipment, the kitchen layout in the self-inspection area image is analyzed to obtain the kitchen area and the kitchen free space; Based on the area of ​​the kitchen area and the free space in the kitchen, a recommended position for placing equipment is generated, and a device image corresponding to the added equipment is generated at the device placement position in the kitchen layout screenshot image, and the kitchen layout screenshot image including the device image is displayed.

[0027] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0028] See also Figure 2 , Figure 2 is a schematic flow chart of a method for generating a rectification strategy based on self-examination content provided in an embodiment of the present application. Figure 2 As shown, the method for generating a rectification strategy based on self-examination content may include steps S101 to S103.

[0029] Step S101: based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, obtain candidate questions corresponding to the target place; This embodiment automatically generates the target rectification strategy of the target site through the strategy generation model, thereby improving the self-inspection efficiency of the target site and the rectification strategy generation efficiency.

[0030] In order to further improve the adaptability of the target place and the rectification strategy, firstly, the personalized self-inspection questions related to the self-inspection area image corresponding to the target place are screened out as candidate questions through the rectification specification knowledge map, and then the target questions most relevant to the target place are determined based on the entropy reduction principle, thereby improving the adaptability of the self-inspection questions and the target place. Then, based on the target questions, interactive questions and answers are conducted with the user, and targeted questions are asked to the user, so as to deeply mine the dynamic description information of the user's self-inspection results of the target place as answer information, and finally the dynamic description information is input into the strategy generation model to obtain the target rectification strategy corresponding to the target place. As a result, not only the labor cost is reduced, the self-inspection efficiency and the generation efficiency of the rectification strategy are improved, but also the self-inspection accuracy and the adaptability of the rectification strategy are improved.

[0031] Specifically, the rectification specification knowledge graph includes non-standard instances, standard instances and rectification information from non-standard to standard corresponding to various standard regulations of various service places. Candidate questions are a set of questions corresponding to the self-inspection specifications related to each self-inspection area in the target place.

[0032] It is understandable that the candidate questions corresponding to the target place can be determined based on the basic business information corresponding to the target place, and the basic business information includes the main business type, business place area, business location, etc.

[0033] A set of question templates corresponding to each self-inspection area and business information can be generated in advance, and various data and information involved in the business activities of each self-inspection area and the target place can be identified based on the self-inspection area image, and candidate questions can be determined in the question template set based on the identified self-inspection areas.

[0034] Specifically, the various data and information involved in the business activities of the target place include the main business format options corresponding to the target place, wherein the specific categories of the main business format include at least one of the following: wholesalers that mainly sell food, shopping malls and supermarkets, convenience stores, specialty stores, concurrent food stores, concurrent drug stores, grocery stores, food traders, Internet food sales operators without physical stores, operators that sell food using automatic equipment, and headquarters of food sales chain enterprises, etc. Catering service operators refer to service providers that provide food and beverages prepared on site and consumed immediately, and the specific categories of the main business format include at least one of the following: large restaurants that mainly prepare and sell food, medium-sized restaurants, small restaurants, central kitchens, collective dining distribution units, pastry shops, beverage shops, small restaurants, snack shops, non-ready-to-eat food preparation and sales stores, operators that prepare and sell food using automatic equipment, headquarters of catering service chain enterprises, catering service management companies, other catering operators, etc. Canteens for centralized dining units usually refer to places set up within schools, enterprises, and institutions to provide collective dining services for students or employees. The specific categories of their main business formats include at least one of the following: school canteens, canteens for childcare institutions, canteens for nursing institutions, canteens for care institutions, canteens for medical institutions, canteens for confinement centers, canteens for childcare institutions, canteens for the elderly, canteens for training institutions, canteens for off-campus care institutions, canteens for government agencies and institutions, enterprise canteens, construction site canteens, and other canteens.

[0035] The production and operation sites can be divided into two major areas: the dining area and the food processing area. The food processing area can be further divided into the receiving area, warehouse area, primary processing area, tableware washing area / room, tableware cleaning area / room, hot food processing area, cold food room, plant-based cold food production area, raw food room, homemade beverage area, pastry processing area, cold processed pastry room, meal preparation area / room, auxiliary area, simple production and sales area, etc. The dining area can be divided into the front door welcoming area, front desk management area, customer dining area, etc.

[0036] Based on the above-mentioned areas, the regulations and norms of each service place are analyzed and summarized, and the problems corresponding to each self-inspection area and the normative instances and non-normative instances corresponding to the problems are obtained as a rectification and normative knowledge map.

[0037] Self-inspection issues include but are not limited to the following: site selection, location and layout, food handling area, cleaning facilities, equipment and tools, food storage, ventilation and lighting, sanitary protection, hand washing and changing, special rooms, special operating areas, water used for food and others.

[0038] Furthermore, the step of obtaining candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user specifically includes: Performing semantic analysis on the self-inspection area image to obtain identifications of each self-inspection area; Based on the self-inspection area identifier, determine the relevant regulations and their corresponding rectification graphic subsets in the rectification regulation knowledge graph; Based on the relevant regulations and their corresponding rectification graphic subsets, the self-inspection area image is analyzed to generate the candidate questions.

[0039] In this embodiment, non-standard instances and standardized instances of various self-inspection projects of various types of enterprises, as well as the rectification process of the same self-inspection project from non-standard to standardized, are collected, and a comparison chart before and after the rectification is displayed in a graphic form, and corresponding rectification instructions for regulatory regulations are provided to establish a rectification instruction database; based on the self-inspection results of the current enterprise, the regulatory regulations that do not meet the standards are determined, and the rectification graphic subset corresponding to the regulatory regulations is queried from the rectification instruction database, and then the most suitable rectification instance is matched in combination with the actual self-inspection project situation of the current enterprise (such as layout, enterprise size, space, etc.), and a rectification guidance strategy is generated as an example, or a guidance strategy is generated for some areas.

[0040] Furthermore, the problem template set corresponding to each self-inspection area and business information is pre-generated, including: Obtaining the food safety laws and regulations and internal management systems, and identifying the inspection items in the food safety laws and regulations and internal management systems that are associated with the self-inspection area and business information; A plurality of verification sections and self-examination questions corresponding to each of the verification sections are generated according to the verification items to obtain a question template set.

[0041] For example, if the check item is a special room, a special room section is generated, which includes at least one self-check question related to the special room, such as whether the special room has an open ditch, whether the floor drain has a water seal, whether the wall skirt is laid to the top, whether it is equipped with an automatically closing special room door, whether it is equipped with an open and close food delivery window, whether it is equipped with an independent air conditioner, whether it is equipped with a refrigerator, whether it is equipped with an ultraviolet lamp, whether the entrance is equipped with a wash basin, whether it is equipped with a tool washing basin, etc. Similarly, there is no fixed order for displaying the above self-check contents.

[0042] Questions for business information may include: "Do you sell hot food?", "Do you sell cold food?", "Do you sell raw food?", "Do you make your own drinks?" Step S102: determining a target question corresponding to the target place from the candidate questions based on the entropy reduction principle, and performing an interactive self-examination inquiry to the user based on the target question to obtain the user's answer information; In this embodiment, based on the keywords in the question and the basic business information of the target place, the correlation degree between the target place and each candidate question is calculated, that is, based on the target place and each candidate question, the entropy value corresponding to the target place and each candidate question is calculated.

[0043] The entropy reduction principle is: the larger the entropy value, the higher the degree of association, and the smaller the entropy value, the lower the degree of association. Based on the order from high to low, a preset number of questions are screened out from the candidate questions, or questions with entropy values ​​exceeding the preset value are screened out as target questions, thereby reducing the target questions associated with the target place and improving the degree of association between the target questions and the target place.

[0044] Based on the selected target questions, interactive self-query questions are performed to users: Specifically, multiple target questions can be displayed in the verification section interface; Exemplarily, the questions and options corresponding to the check items may be displayed on the check section interface; Displaying each target question on the self-check interface corresponding to the verification section, and displaying the self-check content, the standard example diagram of the self-check content, and the answer options for the self-check content on the self-check interface for the user to select; According to the user's selection operation of the answer option for the self-examination content, the self-examination question corresponding to the next self-examination content, the standard example diagram of the next self-examination content, and the answer option for the next self-examination content are displayed.

[0045] Therefore, by intuitively displaying the self-inspection areas and self-inspection questions corresponding to the business information and standard example diagrams on the self-inspection interface, operators do not need to use professional terms to judge the food they operate, thereby determining their business items more quickly and efficiently.

[0046] Step S103: Generate a target rectification strategy for the target place based on a preset strategy generation model, basic business information and the answer information, so that the user can rectify the target place based on the target rectification strategy.

[0047] In this embodiment, the strategy generation model is trained by self-checking problems and their corresponding non-standard instances, standard instances, and rectification information from non-standard to standard.

[0048] The static basic business information and the dynamic information according to the user's answer information are input into the strategy generation model to obtain the target rectification strategy of the target place. The target rectification strategy is displayed to the user so that the user can make corresponding rectifications to the problems existing in the target place based on the target rectification strategy.

[0049] This embodiment provides a method for generating a rectification strategy based on self-examination content, by obtaining candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-examination area image of the target place uploaded by the user; based on the entropy reduction principle, the target question corresponding to the target place is determined in the candidate questions, and an interactive self-examination inquiry is performed to the user based on the target question to obtain the user's answer information; based on the preset strategy generation model, basic business information and the answer information, a target rectification strategy for the target place is generated so that the user can rectify the target place based on the target rectification strategy. Through the above method, the present application first screens out personalized self-examination questions related to the self-examination area image through the rectification specification knowledge graph as candidate questions, thereby improving the adaptability of the self-examination questions to the target place, and then determines the target question most relevant to the target place based on the entropy reduction principle. Then, based on the target question, interactive questions and answers are conducted with the user, and targeted questions are asked to the user, so as to deeply mine the dynamic description information of the user's self-examination results of the target place as answer information, and finally the dynamic description information is input into the strategy generation model to obtain the target rectification strategy corresponding to the target place. As a result, not only the labor cost is reduced and the efficiency of self-inspection and the generation of rectification strategies is improved, but also the accuracy of self-inspection and the adaptability of rectification strategies are improved.

[0050] Furthermore, before generating the target rectification strategy for the target place based on the preset strategy generation model, the basic business information and the answer information, the method further includes: Initialize the model parameters of each network layer in the strategy generation model and generate the probability parameters of each network layer; Based on the probability parameters of each network layer, determine the network layer to be trained and the network layer to be optimized in each network layer; Based on the preset training sample set and the initial model parameters, the network layer to be optimized is skipped, and the network layer to be trained is trained to obtain the strategy generation model.

[0051] Wherein, based on the probability parameters of each network layer, determining the network layer to be trained and the network layer to be optimized in each network layer specifically includes: Obtain any value within a preset value range as a random parameter of a network layer; Compare the probability parameters of each network layer with its corresponding random parameters; Based on the comparison results of each network layer, the network layer whose probability parameter is not less than the random parameter is determined as the network layer to be trained, and the network layer whose probability parameter is less than the random parameter is determined as the network layer to be optimized.

[0052] In this embodiment, in order to prevent the policy generation model from overfitting or underfitting, thereby causing the training time of the policy generation model to be too long or the model accuracy to be low, a random depth mechanism is provided based on the survival probability corresponding to each network layer. Based on the random depth mechanism, some network layers can be randomly skipped during the training process, thereby reducing the amount of forward and backward propagation calculations during the policy generation model training, thereby reducing the computational complexity during the policy generation model training; and only part of the network layers are used in each training, so that the policy generation model can significantly shorten the training time while ensuring the convergence of the policy generation model, preventing the policy generation model from overfitting, and enhancing the robustness of the policy generation model to abnormal data and noise data.

[0053] Specifically, the strategy generation model is a random deep neural network model. By introducing the random depth mechanism in the training process, efficient training of the deep neural network is achieved, including: 1. Initialize the network parameters, that is, initialize the weight parameters and bias parameters of each layer of the strategy generation model. Specifically, the model parameters of the random deep neural network can be initialized by using methods such as He initialization or Xavier initialization to ensure the stability and effectiveness of the training.

[0054] Second, initialize the survival probability. The probability parameter is the survival probability parameter, that is, assign an initial survival probability to each layer of the network .

[0055] Specifically, a linear decreasing strategy can be adopted, that is, the level of the network layer is inversely proportional to the corresponding survival probability parameter value. By adopting a linear decreasing survival probability design, the effective depth of the network during the training process is guaranteed, while avoiding overly complex calculations.

[0056] That is, the survival probability is higher near the input layer, while the survival probability is lower near the output layer. For example, the survival probability can be set as:

[0057] in, Less than 1, is the current network layer, is the total number of network layers, It is the survival probability of the output layer in the previous round. The initial value can be taken in the first round. The value is generally low, such as less than 50%.

[0058] A preset probability threshold can be set. When the survival probability parameter is not less than the preset probability threshold, the network layer can be recorded as the target network layer to train the network layer. When the survival probability parameter is less than the preset probability threshold, the network layer can be recorded as the optimized network layer to skip the network layer.

[0059] In a specific embodiment, a corresponding random number may be generated for each network layer, and the survival probability parameter of each network layer may be compared with the corresponding random number. When the survival probability parameter is not less than the random number, the network layer may be recorded as the target network layer; when the survival probability parameter is less than the random number, the network layer may be recorded as the optimized network layer.

[0060] The initial random deep network model is a random deep network model with initial weight parameters and initial bias parameters as the current model parameters. In each forward propagation process, the business description information in the preset strategy training sample set is input into each target network layer in the initial random deep network model, that is, the output results of each target network layer skip the next layer of the optimization network layer and directly pass it to the next layer of the target network layer. After completing the training of all target network layers in the random deep network model, it is determined whether the current random deep network model has reached the convergence condition. If not, the target network layer in the random deep network model is iteratively trained again, and the optimization network layer is still skipped until the random deep network model reaches the preset convergence condition.

[0061] Among them, the preset convergence condition can be that the loss function of the random deep network model converges (that is, the loss function value is less than a preset function threshold), or the number of training times of the random deep network model reaches a preset number of training rounds.

[0062] Specifically, the business description information and its corresponding rectification strategy information are first input into the first-level network layer.

[0063] Then, each network layer except the first-level network layer in the random deep network model is processed layer by layer, and each network layer of the initialized random deep network model is forward propagated: For each network layer , generate a random number between [0, 1] , as random parameters of each level of network layer; The random parameters of each network layer The corresponding survival probability parameter To compare: 1. If , then this layer participates in the calculation and performs normal forward propagation:

[0064] in, For the The input of the layer, and Respectively The weights and biases of the layers, is the activation function of this layer.

[0065] 2. If , then skip this layer and pass the current input directly to the next layer:

[0066] In the above manner, this embodiment generates corresponding random parameters for each level of network layer to ensure the randomness of skipping the layer, and then compares the random parameters of the layer with its corresponding survival probability parameters, and determines whether to skip the layer during training based on the comparison results. Thus, the randomness of the optimized network layer is ensured based on the random parameters, thereby avoiding the fixed probability parameter threshold leading to a fixed optimized network layer, and further improving the accuracy of the prediction results of the random deep network model.

[0067] After completing a round of forward propagation of the model, back propagation is performed to update the parameters of the network layers involved in the forward propagation. Specifically: 1. Calculate the target rectification strategy (i.e. target value) corresponding to the business description information sample ) and the current rectification strategy (network output )’s loss function: According to the network output and target value Calculating the loss function , using cross entropy loss or mean squared error loss.

[0068] 2. Back propagation layer by layer: Starting from the output layer, the gradient is calculated forward layer by layer.

[0069] For each layer , if the layer participates in the forward propagation (i.e. ), the gradient is calculated normally and the parameters are updated:

[0070]

[0071] in, is the preset learning rate. If the layer is skipped (i.e. ), no parameter update is performed and the gradient is directly passed to the previous layer:

[0072] In the above manner, this embodiment will iterate multiple times on the entire training set based on the above forward propagation and back propagation steps until the loss function converges or reaches a preset number of training rounds. During each iteration, the survival probability of each network layer in this round is Keep it unchanged to ensure the stability and effect of model training.

[0073] After completing the training of the random depth network model, the model is further tested to ensure the stability and accuracy of the model's prediction results. The test methods include full depth test method (i.e., no network layer is skipped during the test) and random depth test method (i.e., some network layers are skipped based on survival probability). Specifically: A. The full-depth test uses a full-depth network for forward propagation based on the test sample set, that is, no network layer in the target random depth network model is skipped during the test to obtain the test accuracy of the target random depth network model.

[0074] B. Random depth testing uses a random depth mechanism to perform forward propagation based on a test sample set. That is, during the test process, some network layers are skipped with the same or different survival probabilities to simulate the actual situation during training and obtain the test accuracy.

[0075] When the test accuracy is less than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are high; when the test accuracy is greater than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are low. The target random deep network model is updated until the test accuracy is less than the preset accuracy threshold to ensure the stability and accuracy of the prediction results of the random deep network model.

[0076] The embedding layer and convolution layer in the random deep training algorithm are used to reduce the dimensionality and extract features of the high-dimensional data corresponding to the sample set, thereby effectively reducing the computational complexity and training time of the deep neural network. While improving data processing efficiency and prediction accuracy, it further improves the robustness and performance of the random deep network model.

[0077] Furthermore, before obtaining candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, the method further includes: Collect project information corresponding to various regulations and rules of various places, wherein the project information includes non-standard examples, standard examples, and rectification information from non-standard to standard; Segment the project information into at least one text unit, and extract at least one entity information and relationship information corresponding to each entity from each text unit based on a language model; Based on the entity information and the relationship information corresponding to each entity, the rectification specification knowledge graph is generated.

[0078] Furthermore, generating the rectification specification knowledge graph based on each entity information and the relationship information corresponding to each entity specifically includes: Based on each of the entity information and the relationship information corresponding to each entity, construct a weighted undirected graph corresponding to the project information, wherein the nodes of the weighted undirected graph represent the entity information, and the edges of the weighted undirected graph represent the relationship information; The weighted undirected graph is hierarchically clustered to obtain information of each community, and based on the summary corresponding to each community, the rectification specification knowledge graph is generated.

[0079] In this embodiment, the GraphRAG indexing algorithm is used to construct a rectification standard knowledge graph. First, data such as non-standardized instances, standardized instances, and rectification information from non-standardized to standardized instances of various places are collected, and then the above data is preprocessed, including but not limited to text cleaning, removal of irrelevant information, and terminology standardization. Specifically, the following steps are included: 1. Text unit segmentation: Segment the preprocessed project information to obtain basic text units that are convenient for subsequent analysis as text units (TextUnits).

[0080] 2. Entity and relationship extraction: Use the Large Language Model (LLM) to extract all entities (such as special rooms, open drains, floor drains, water seals, wall skirts, special room doors, food delivery windows, independent air conditioners, refrigerators, UV lamps, wash basins, tool cleaning pools, etc.), relationships (such as the corresponding questions of each self-inspection area and the corresponding relationship between the standard instances and non-standard instances of the questions) and key statements from each text unit. By optimizing prompts and multiple rounds of collection, we can ensure that the extraction results are of high quality even when processing large blocks of text.

[0081] 3. Knowledge graph construction and community segmentation: Use the extracted entities and relationships to build a homogeneous weighted undirected graph. Each node in the graph is used to represent each entity extracted above, the edge represents the relationship, and the weight of the edge represents the standardized count of the detected relationship instance.

[0082] The Leiden algorithm is applied to perform hierarchical clustering on the knowledge graph to form different communities. The Leiden algorithm optimizes the traditional Louvain algorithm to run faster and ensure connectivity within the community, making the community division more reasonable.

[0083] 4. Generate community summaries: For each community, generate summaries of the community and its members from the bottom up. Create a report-style summary for each community in the Leiden hierarchy. In order to understand the global structure and semantics of the dataset through the community summary.

[0084] 5. Store the knowledge graph: Store the constructed knowledge graph in the graph database Neo4j to facilitate subsequent query and analysis.

[0085] Furthermore, after generating the target rectification strategy of the target place based on the preset strategy generation model, the basic business information and the answer information, it also includes: When the target rectification strategy is to add necessary equipment, the kitchen layout in the self-inspection area image is analyzed to obtain the kitchen area and the kitchen free space; Based on the area of ​​the kitchen area and the free space in the kitchen, a recommended position for placing equipment is generated, and a device image corresponding to the added equipment is generated at the device placement position in the kitchen layout screenshot image, and the kitchen layout screenshot image including the device image is displayed.

[0086] In this embodiment, the video or picture of the target place uploaded by the user can be automatically analyzed to obtain the current user's business environment, and the user's answer information can be obtained based on the target question. When it is determined that the target place does not meet the requirements based on the answer information, rectification suggestions are generated.

[0087] Specifically, rectification suggestions can be presented to users in the form of pictures, that is, the areas with problems in the video are captured, and the existing problems and rectification suggestions are marked. For example, if there is no problem with the equipment in the current kitchen but some necessary equipment is missing, then pictures of the areas where the equipment can be placed are captured according to the kitchen layout. If the inspection result shows that there is currently a lack of a disinfection cabinet, the kitchen layout can be analyzed based on the kitchen video uploaded by the merchant, and the location of the disinfection cabinet can be intelligently recommended based on the kitchen layout, and the smart cabinet can be displayed in the picture corresponding to the location so that the merchant can view the specific location of the smart cabinet; the merchant can also intelligently recommend a disinfection cabinet of appropriate size to the merchant based on the kitchen size and free space obtained from the video analysis to save kitchen space. (Rectification suggestions can be obtained from the business plans of historical merchants that meet the business specifications to meet the current user requirements).

[0088] It is understandable that the user in this embodiment is a store operator. The type of business premises and the location of the business premises of the target place can be obtained based on the store logo. The store logo can be obtained directly or indirectly. Direct acquisition is based on the unique identification information provided by the store for identification, which can be directly input by the user or identified by uploaded documents. Indirect acquisition is determined through intelligent question and answer with the user. When directly acquired, it can be further verified through the Internet to further determine the correctness of the information. Indirect acquisition can facilitate users of unregistered stores to obtain and authenticate store information, reducing food safety risks caused by unregistered or unlicensed stores in society.

[0089] In some embodiments, before obtaining the food business information of the operator through the above indirect method, the method further includes: Display a welcome interface, wherein the welcome interface displays a welcome message and a prompt for obtaining food business information, wherein the prompt for obtaining food business information is used to prompt the operator to enter the unique identification information of the enterprise, wherein the unique identification information of the enterprise includes at least one of the following: a unified social credit code, a business license number, an organization code, a tax registration number, a bank account opening license number, and a company name; Obtain the food business license of the enterprise in the background based on the unique identification information of the enterprise entered by the operator; If the acquisition is successful, the main business format and business items will be extracted according to the food business license; If the acquisition fails, the food business information intelligent acquisition option and the license intelligent image recognition option are displayed on the welcome interface; If it is detected that the operator has selected the option of intelligently obtaining the food business information, the program will jump to the main business interface and obtain the food business information of the operator in the above-mentioned indirect manner; If it is detected that the operator has selected the smart image recognition option of the license, the operator is prompted to upload the food business license, and the operator's food business information is extracted based on the food business license uploaded by the operator.

[0090] It can be seen that in the embodiment of the present application, before obtaining the operator's food business information indirectly, it is first determined whether the operator's food business information can be obtained directly, which is more accurate than food business information obtained indirectly; the direct method of obtaining the operator's food business information is first to see whether it is possible to directly obtain the company's unique identification information, and then obtain the company's food business license in the background based on the unique identification information. It is inconvenient for the operator to carry the food business license with him, which can save the operator's time in looking for the food business license and improve work efficiency; when the acquisition fails, the operator's food business information is extracted through the food business license uploaded by the operator, and the operator does not need to check the food business information on the food business license and then manually enter it, which can also improve work efficiency.

[0091] The type of business premises refers to the main business format of the store, which means the type of business or operating model that the store mainly engages in during its operation.

[0092] In some embodiments, the user may be asked about the production and operation site information through intelligent dialogue, and the production and operation site information includes generating an example diagram of the business site layout and equipment layout. The method may include: In the question area of ​​the main business option details display interface, ask the user whether to choose to upload the production and operation site layout diagram; If yes, receiving the layout diagram of the production and operation site uploaded by the user; If not, fill in the blanks for the area of ​​production and operation sites in the question area of ​​the main business option details display interface; Among them, the example diagram of the production and operation site layout and equipment layout can change dynamically with the area input by the operator. If the area is within the area range of micro-catering, the example diagram of the production and operation site layout and equipment layout is the example diagram of the production and operation site layout and equipment layout of micro-catering; if the area is within the area range of small-sized catering, the example diagram of the production and operation site layout and equipment layout is the example diagram of the production and operation site layout and equipment layout of small-sized catering; if the area is within the area range of medium-sized catering, the example diagram of the production and operation site layout and equipment layout is the example diagram of the production and operation site layout and equipment layout of medium-sized catering. The example diagram of the production and operation site layout and equipment layout of the corresponding surface is displayed in real time through dynamic changes, which is convenient for operators to compare, determine and adjust, and find a place that suits their actual operation, so that the information of the production and operation site is more accurate.

[0093] It can be seen that in the embodiment of the present application, even if the user does not upload a layout map of the production and operation site, an approximate layout map of the production and operation site can be generated through intelligent dialogue, which is conducive to the subsequent self-inspection area division, rather than conducting self-inspection of all areas in general, regardless of whether the area exists in the production and operation site.

[0094] Furthermore, when receiving the answer information of the current user to each self-examination question, diagnosing each self-examination question based on each self-examination standard text and each corresponding standard image instance, and generating a diagnosis report corresponding to each self-examination area, it also includes: Collecting non-standard instances of each self-inspection area corresponding to each type of store, and collecting rectification process information of each type of store from non-standard to standard self-inspection area, wherein the rectification process information includes comparison pictures before and after rectification; Based on the comparison charts before and after the rectification, answer options corresponding to each self-examination question are generated, so that the current user can make selections based on the situation of his own store to generate corresponding answer information.

[0095] In this embodiment, the answer options may also be binary options such as "yes" or "no", or may be an input box, which may be used to input data such as text, voice, picture or video.

[0096] The various rectification information in the target rectification strategy can also be further displayed based on the inspection order. Among them, the inspection order can be the inspection order of each self-inspection area in the best inspection path. The best inspection path can be based on the path calculation of the global path planning algorithm. All path planning algorithms include RRT algorithm*, PRM algorithm and Theta* algorithm. The specific planning algorithm can be determined based on the operator's selection operation or the site characteristics of the target site.

[0097] Furthermore, if Figure 3 As shown, when it is determined that multiple verification sections have completed the verification based on the self-inspection completion degree, a diagnosis is performed and a self-inspection diagnosis report is generated based on the operator's selection operation of the answer option to the target question (i.e., the self-inspection question) corresponding to each verification section.

[0098] The degree of self-check completion can be displayed in the form of a progress bar, a percentage, a score, or other forms of display, which are not listed here. Specifically, according to the degree of self-check completion, it can be determined that multiple check blocks have completed the check by displaying all progress bars, or by determining that multiple check blocks have completed the check when the score is 1 or the percentage is 100%.

[0099] In an embodiment of the present application, when it is determined based on the degree of self-inspection completion that multiple inspection sections have been completed, a diagnosis is performed and a diagnostic report is generated based on the operator's selection operation of the answer options to the self-inspection content corresponding to each inspection section, and no diagnosis is performed or a diagnostic report is generated when there are incomplete inspection sections. This can achieve the purpose of reminding the operator to conduct further self-inspection and prevent the generated diagnostic report from being incomplete, which may cause the operator to mistakenly believe that he has completed all self-inspections and achieved qualified results, thereby increasing food safety risks.

[0100] In order to better illustrate the effect of the embodiment of the present application, the target problem is taken as an example of a wall skirt. When the current self-inspection content is whether the wall skirt is laid to the top, when the operator's self-inspection content is whether the entrance is equipped with a wash basin and whether it is equipped with a tool cleaning pool, the actual business map uploaded can at least partially reflect whether the wall skirt is laid to the top. Therefore, when the actual business map uploaded by the operator for the self-inspection content of whether the wall skirt is laid to the top cannot determine whether the self-inspection content is in compliance, further judgment is made through the actual business map uploaded by the operator for the self-inspection content of whether the entrance is equipped with a wash basin and whether it is equipped with a tool cleaning pool, so as to achieve the purpose of diagnosis.

[0101] In some embodiments, the answer options include binary confirmation options of "yes" and "no" and a selection input box, and the method further includes: When the user's answer option to the self-examination content is "yes" to confirm the option, prompt the operator to input the actual business diagram in the selection input box; Receiving the actual business diagram uploaded by the user for the standard example diagram of the self-examination content, wherein the actual business diagram refers to a real picture or diagram taken or drawn by the operator (i.e., the user) during the actual business process; The step of performing diagnosis and generating a diagnosis report according to the user's selection operation of answering options to the self-examination content includes: A diagnosis is performed based on the actual operation diagram and a diagnosis report is generated, wherein the diagnosis report refers to a written report formed after a comprehensive inspection and evaluation of the target site.

[0102] Furthermore, an embodiment of the present application provides a schematic diagram of a structure of a device for generating a rectification strategy based on self-examination content, the device comprising: A candidate question generation module, used to obtain candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user; A question interactive answering module, configured to determine a target question corresponding to the target place from the candidate questions based on the entropy reduction principle, and perform an interactive self-examination inquiry to the user based on the target question to obtain the user's answer information; The rectification strategy generation module is used to generate a target rectification strategy for the target place based on a preset strategy generation model, basic business information and the answer information, so that the user can rectify the target place based on the target rectification strategy.

[0103] Furthermore, the device also includes: A probability parameter generation module is used to initialize the model parameters of each network layer in the strategy generation model and generate probability parameters of each network layer; A network layer classification module, used for determining a network layer to be trained and a network layer to be optimized in each network layer based on probability parameters of each network layer; The strategy model generation module is used to skip the network layer to be optimized based on a preset training sample set and initial model parameters, and train the network layer to be trained to obtain the strategy generation model.

[0104] Furthermore, the network layer classification module specifically includes: A random parameter generating unit, used to obtain any value within a preset value range as a random parameter of a network layer; A network parameter comparison unit, used to compare the probability parameters of each network layer with its corresponding random parameters; The network layer classification unit is used to determine, based on the comparison results of each network layer, that the network layer whose probability parameter is not less than the random parameter is the network layer to be trained, and to determine that the network layer whose probability parameter is less than the random parameter is the network layer to be optimized.

[0105] Furthermore, the device also includes: The project information collection module is used to collect project information corresponding to various regulations and rules of various places, wherein the project information includes non-standard examples, standard examples and rectification information from non-standard to standard; A text segmentation and extraction module, used to segment the project information into at least one text unit, and extract at least one entity information and relationship information corresponding to each entity from each text unit based on a language model; The knowledge graph generation module is used to generate the rectification specification knowledge graph based on the entity information and the relationship information corresponding to each entity.

[0106] Furthermore, the knowledge graph generation module specifically includes: An undirected graph construction unit, configured to construct a weighted undirected graph corresponding to the project information based on each of the entity information and the relationship information corresponding to each of the entities, wherein the nodes of the weighted undirected graph represent the entity information, and the edges of the weighted undirected graph represent the relationship information; The knowledge graph construction unit is used to perform hierarchical clustering on the weighted undirected graph to obtain information of each community, and generate the rectification specification knowledge graph based on the summary corresponding to each community.

[0107] Furthermore, the candidate question generating module specifically includes: A region identification generating unit, configured to perform semantic analysis on the self-inspection region image to obtain each self-inspection region identification; A rectification graphic and text generation unit, used to determine relevant regulations and their corresponding rectification graphic and text subsets in the rectification specification knowledge graph based on the self-inspection area identifier; The candidate question generating unit is used to analyze the self-inspection area image based on the relevant regulations and their corresponding rectification graphic subsets to generate the candidate questions.

[0108] Furthermore, the device also includes: A regional layout analysis module, used to analyze the kitchen layout in the self-inspection area image to obtain the kitchen area and the kitchen vacant positions when the target rectification strategy is to add necessary equipment; A strategy image display module is used to generate recommended equipment placement positions based on the area of ​​the kitchen area and the vacant positions in the kitchen, generate equipment images corresponding to the equipment placed at the equipment placement positions in the kitchen layout screenshot image, and display the kitchen layout screenshot image including the equipment images.

[0109] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0110] A computer-readable storage medium is also provided in an embodiment of the present application. The computer-readable storage medium stores a computer program. The computer program includes program instructions. The processor executes the program instructions to implement any one of the rectification strategy generation methods based on self-inspection content provided in the embodiments of the present application.

[0111] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0112] The computer-readable storage medium may be an internal storage unit of the computer device of the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital card (Secure Digital, SD), a flash card (Flash Card), etc. equipped on the computer device.

[0113] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for generating a rectification strategy based on self-examination content, characterized in that: The method comprises: Based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, the candidate questions corresponding to the target place are obtained; Based on the entropy reduction principle, determining a target question corresponding to the target place from the candidate questions, and performing an interactive self-examination inquiry to the user based on the target question to obtain answer information of the user; Based on the preset strategy generation model, basic business information and the answer information, a target rectification strategy for the target place is generated so that the user can rectify the target place based on the target rectification strategy.

2. The method for generating a rectification strategy based on self-examination content according to claim 1, characterized in that: Before generating the target rectification strategy of the target place based on the preset strategy generation model, the basic business information and the answer information, the method further includes: Initialize the model parameters of each network layer in the strategy generation model and generate the probability parameters of each network layer; Based on the probability parameters of each network layer, determine the network layer to be trained and the network layer to be optimized in each network layer; Based on the preset training sample set and the initial model parameters, the network layer to be optimized is skipped, and the network layer to be trained is trained to obtain the strategy generation model.

3. The method for generating a rectification strategy based on self-examination content according to claim 2, characterized in that: The determining of the network layer to be trained and the network layer to be optimized in each network layer based on the probability parameters of each network layer specifically includes: Obtain any value within a preset value range as a random parameter of a network layer; Compare the probability parameters of each network layer with its corresponding random parameters; Based on the comparison results of each network layer, the network layer whose probability parameter is not less than the random parameter is determined as the network layer to be trained, and the network layer whose probability parameter is less than the random parameter is determined as the network layer to be optimized.

4. The method for generating a rectification strategy based on self-examination content according to claim 1, characterized in that: Before obtaining candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user, the method further includes: Collect project information corresponding to various regulations and rules of various places, wherein the project information includes non-standard examples, standard examples, and rectification information from non-standard to standard; Segment the project information into at least one text unit, and extract at least one entity information and relationship information corresponding to each entity from each text unit based on a language model; Based on the entity information and the relationship information corresponding to each entity, the rectification specification knowledge graph is generated.

5. The method for generating a rectification strategy based on self-examination content according to claim 4, characterized in that: The step of generating the rectification specification knowledge graph based on the entity information and the relationship information corresponding to each entity specifically includes: Based on each of the entity information and the relationship information corresponding to each entity, construct a weighted undirected graph corresponding to the project information, wherein the nodes of the weighted undirected graph represent the entity information, and the edges of the weighted undirected graph represent the relationship information; The weighted undirected graph is hierarchically clustered to obtain information of each community, and based on the summary corresponding to each community, the rectification specification knowledge graph is generated.

6. The method for generating rectification strategies based on self-examination content according to claim 1, characterized in that: The step of obtaining candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user specifically includes: Performing semantic analysis on the self-inspection area image to obtain identifications of each self-inspection area; Based on the self-inspection area identifier, determine the relevant regulations and their corresponding rectification graphic subsets in the rectification regulation knowledge graph; Based on the relevant regulations and their corresponding rectification graphic subsets, the self-inspection area image is analyzed to generate the candidate questions.

7. The method for generating a rectification strategy based on self-examination content according to any one of claims 1 to 6, characterized in that: After generating the target rectification strategy of the target place based on the preset strategy generation model, the basic business information and the answer information, the method further includes: When the target rectification strategy is to add necessary equipment, the kitchen layout in the self-inspection area image is analyzed to obtain the kitchen area and the kitchen free space; Based on the area of ​​the kitchen area and the free space in the kitchen, a recommended position for placing equipment is generated, and a device image corresponding to the added equipment is generated at the device placement position in the kitchen layout screenshot image, and the kitchen layout screenshot image including the device image is displayed.

8. A device for generating rectification strategies based on self-examination content, characterized in that: The device comprises: A candidate question generation module, used to obtain candidate questions corresponding to the target place based on the rectification specification knowledge graph and the self-inspection area image of the target place uploaded by the user; A question interactive answering module, configured to determine a target question corresponding to the target place from the candidate questions based on the entropy reduction principle, and perform an interactive self-examination inquiry to the user based on the target question to obtain the user's answer information; The rectification strategy generation module is used to generate a target rectification strategy for the target place based on a preset strategy generation model, basic business information and the answer information, so that the user can rectify the target place based on the target rectification strategy.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the method for generating a rectification strategy based on self-inspection content as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for generating a rectification strategy based on self-inspection content according to any one of claims 1 to 7 is implemented.

Images