A hotel service push method, device and equipment based on big data

Abstract

The present application provides a hotel service push method, device and equipment based on big data. First, two demand analysis models are obtained by pre-adjusting a hotel service demand analysis model according to an active type of behavior data template and a behavior data template of a type to be classified, so that the two demand analysis models generate certain demand analysis performances, which helps to accurately obtain a fuzzy index and accurately optimize a prior label. By comparing the demand analysis values obtained by the two demand analysis models for the same behavior data template of the type to be classified, the fuzzy index of the behavior data template of the type to be classified is determined based on the error, so as to determine the importance of the behavior data template of the type to be classified in subsequent iterative debugging, and mitigate the disturbance information caused by low-quality behavior data templates that are not suitable for participating in debugging. Based on the demand analysis value, the prior label of some behavior data templates of the type to be classified is adjusted to complete the correction, and the disturbance information of the prior label of the behavior data templates of each type to be classified is mitigated.

Description

Technical Field

[0001] This application relates to the fields of big data and machine learning, and more particularly, to a hotel service push method, device, and equipment based on big data. Background Art

[0002] In the wave of intelligent services driven by big data, machine learning models are widely used in various industries to achieve accurate identification of user needs and efficient services. As an important part of the service industry, the hotel service industry is also seeking ways to improve service quality and user experience through machine learning technology. However, in practical applications, machine learning models face some unique challenges in the hotel service field.

[0003] First, the user behavior data in the hotel service field is extremely complex and diverse, covering all aspects from reservation to check-in, consumption to check-out, etc. These data contain the real needs and service preferences of users, but there is also a large amount of noise and uncertainty. Traditional machine learning models often have difficulty accurately distinguishing effective information from noise when processing these data, resulting in limited model performance. Second, during the training process of machine learning models, a large number of positive samples (i.e., positive type of behavior data) are usually required to ensure the accuracy of the models. However, in the hotel service field, positive samples are often difficult to obtain or limited in quantity, while negative samples (i.e., behavior data of the type to be classified, which may contain noise) are abundant. This causes the models to be easily affected by the noise in the negative samples during training, reducing the generalization ability and accuracy of the models. As a result, the hotel service needs identified by the debugged models are inaccurate, and the push conversion rate is low. Summary of the Invention

[0004] In view of this, at least one embodiment of this application provides a hotel service push method, device, and equipment based on big data.

[0005] According to one aspect of the embodiments of this application, a hotel service push method based on big data is provided, which is applied to a computer device. The method includes:

[0006] Obtain the target user behavior big data, and load the target user behavior big data into a hotel service demand analysis model that has been debugged in advance, where the target user behavior big data includes the hotel interaction behavior data of the target user;

[0007] Analyze the target user behavior big data according to the hotel service demand analysis model to obtain the hotel service demand of the target user;

[0008] According to the hotel service requirements of the target user, determine the target hotel service that matches the hotel service requirements among the preset hotel services, and push the target hotel service to the client logged in by the target user;

[0009] Among them, the hotel service requirement analysis model is determined based on the behavior data template library. The behavior data template library includes a positive type behavior data template library and a to-be-classified type behavior data template library. The hotel service requirement analysis model is obtained through the following steps of debugging:

[0010] Configure the prior label of each positive type behavior data template as the positive type, configure the prior label of each to-be-classified type behavior data template as the negative type, and based on the prior label of each behavior data template, use the behavior data template library as a training batch to iteratively debug the hotel service requirement analysis model to obtain the first requirement analysis model and the requirement analysis values inferred by the first requirement analysis model for each to-be-classified type behavior data template, as well as the second requirement analysis model and the requirement analysis values inferred by the second requirement analysis model for each to-be-classified type behavior data template. Among them, the first requirement analysis model is obtained by using the first number of iterative debuggings, the second requirement analysis model is obtained by using the second number of iterative debuggings, the first number and the second number are not equal, and in the iterative debugging, optimize the model parameters of the hotel service requirement analysis model based on the requirements determined from the behavior data template library;

[0011] Determine the fuzzy index of each to-be-classified type behavior data template based on the error between the requirement analysis values inferred by the first requirement analysis model and the second requirement analysis model for each to-be-classified type behavior data template in the to-be-classified type behavior data template library;

[0012] Adjust the prior label of each to-be-classified type behavior data template based on the requirement analysis value inferred by the first requirement analysis model for each to-be-classified type behavior data template;

[0013] According to the fuzzy index and the adjusted prior label of each to-be-classified type behavior data template, use the to-be-classified type behavior data template library as a training batch to iteratively debug the first requirement analysis model to obtain the debugged first requirement analysis model;

[0014] Determine the model parameters of the hotel service requirement analysis model according to the model parameters of the debugged first requirement analysis model.

[0015] According to an example of an embodiment of the present application, wherein adjusting the prior label of the behavior data template of each type to be classified according to the requirement analysis value inferred for the behavior data template of each type to be classified based on the first requirement analysis model includes:

[0016] For each behavior data template of each type to be classified in the behavior data template library of the type to be classified, perform the following operations:

[0017] When the requirement analysis value inferred by the first requirement analysis model for the behavior data template of the type to be classified is greater than the positive type critical value, adjust the prior label of the behavior data template of the type to be classified to the positive type;

[0018] When the requirement analysis value inferred by the first requirement analysis model for the behavior data template of the type to be classified is not greater than the positive type critical value, adjust the prior label of the behavior data template of the type to be classified to the negative type.

[0019] According to an example of an embodiment of the present application, wherein iteratively debugging the first requirement analysis model with the behavior data template library of each type to be classified as a training batch according to the fuzzy index sum and the adjusted prior label of the behavior data template of each type to be classified includes:

[0020] Determine the importance coefficient of the behavior data template of each type to be classified based on the fuzzy index of the behavior data template of each type to be classified, and the change trend of the importance coefficient is opposite to that of the fuzzy index;

[0021] Repeatedly perform the following operations until the first requirement analysis model reaches the debugging cut-off condition:

[0022] Determine the requirement determination value of the behavior data template of each type to be classified based on the first requirement analysis model;

[0023] Determine the requirement determination cost of the behavior data template of each type to be classified based on the prior label and the requirement determination value of the behavior data template of each type to be classified;

[0024] Determine the requirement determination cost of the behavior data template library of the type to be classified according to the requirement determination cost and the importance coefficient of the behavior data template of each type to be classified;

[0025] Optimize the model parameters of the first requirement analysis model based on the requirement determination cost of the behavior data template library of the type to be classified.

[0026] According to another aspect of the embodiments of the present application, there is provided a hotel service push device based on big data, including:

[0027] A data acquisition module, configured to acquire big data of target user behaviors, and load the big data of target user behaviors into a hotel service demand analysis model that has been pre-debugged, wherein the big data of target user behaviors includes hotel interaction behavior data of a target user;

[0028] A model calling module, configured to analyze the big data of target user behaviors according to the hotel service demand analysis model to obtain the hotel service demands of the target user;

[0029] A service push module, configured to determine a target hotel service that matches the hotel service demand from preset hotel services according to the hotel service demand of the target user, and push the target hotel service to the client logged in by the target user;

[0030] A model debugging module, configured to debug the hotel service demand analysis model, wherein the hotel service demand analysis model is determined according to a behavior data template library, the behavior data template library includes a behavior data template library of positive types and a behavior data template library of types to be classified, and the hotel service demand analysis model is obtained through the following steps of debugging:

[0031] Configure the prior label of each behavior data template of positive types as positive type, configure the prior label of each behavior data template of types to be classified as negative type, and iteratively debug the hotel service demand analysis model with the behavior data template library as a training batch according to the prior label of each behavior data template, to obtain a first demand analysis model and the demand analysis values inferred by the first demand analysis model for each behavior data template of types to be classified, as well as a second demand analysis model and the demand analysis values inferred by the second demand analysis model for each behavior data template of types to be classified, wherein the first demand analysis model is obtained by using a first number of iterations of debugging, the second demand analysis model is obtained by using a second number of iterations of debugging, the first number is not equal to the second number, and optimize the model parameters of the hotel service demand analysis model based on the demand determination cost of the behavior data template library during the iterative debugging;

[0032] Determine the fuzzy index of each behavior data template of types to be classified based on the error between the demand analysis values inferred by the first demand analysis model and the second demand analysis model for each behavior data template of types to be classified in the behavior data template library of types to be classified;

[0033] Adjust the prior label of each behavior data template of types to be classified based on the demand analysis value inferred by the first demand analysis model for each behavior data template of types to be classified;

[0034] According to the fuzzy index of the behavior data template of each type to be classified and the adjusted prior label, using the behavior data template library of the type to be classified as a training batch to iteratively debug the first demand analysis model, and obtaining the debugged first demand analysis model;

[0035] According to the model parameter variables of the debugged first demand analysis model, determine the model parameter variables of the hotel service demand analysis model.

[0036] According to another aspect of the embodiments of the present application, there is provided a computer device, including: a processor; and a memory, wherein a computer-readable storage medium is stored in the memory, and when the computer-readable storage medium is run by the processor, the processor is caused to execute the method as described above.

[0037] The beneficial effects included in the present application at least:

[0038] The method for pushing hotel services based on big data provided by the embodiments of the present application first pre-debugs the hotel service demand analysis model according to the behavior data templates of positive types and the behavior data templates of types to be classified to obtain two demand analysis models, enabling the two demand analysis models to have certain demand analysis performance, which can help accurately obtain the fuzzy index and accurately optimize the prior label. Further, by comparing the demand analysis values obtained by the two demand analysis models for the behavior data template of the same type to be classified, the fuzzy index of the behavior data template of the type to be classified is determined based on their errors, so as to determine the importance of the behavior data template of the type to be classified in subsequent iterative debugging, and alleviate the disturbance information caused by low-quality behavior data templates that are not suitable for participating in debugging. Furthermore, the prior label of some behavior data templates of the types to be classified is adjusted based on the demand analysis value to complete the correction of the prior label, and alleviate the disturbance information in the prior labels of the behavior data templates of each type to be classified.

[0039] In addition, in the subsequent iterative debugging of the first requirements analysis model, the requirement determination cost of the behavior data templates of each type to be classified is optimized based on the fuzzy index and the adjusted prior tags of the behavior data templates of each type to be classified. Among them, based on the fuzzy index, the influence of the behavior data templates of the types to be classified with larger fuzzy indexes on the optimization of the model parameters of the first requirements analysis model can be restricted. Then, the influence of the behavior data templates of the types to be classified with smaller fuzzy indexes on the optimization of the model parameters of the first requirements analysis model will be highlighted. Combining the adjusted prior tags, since the perturbation information in the prior tags of the behavior data templates of each type to be classified and the perturbation information of the low-quality behavior data templates unsuitable for debugging are restricted, this can improve the requirements analysis effect of the first requirements analysis model. Based on the first requirements analysis model, a hotel service requirements analysis model is obtained. Since the requirements analysis effect of the first requirements analysis model is enhanced, the hotel service requirements analysis model also has excellent requirements analysis performance.

[0040] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the technical solutions of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0042] Figure 1 is a schematic diagram of the architecture of an application scenario provided by this application;

[0043] Figure 2 is a schematic flowchart of a method for pushing hotel services based on big data provided by this application;

[0044] Figure 3 is a schematic structural diagram of a device for pushing hotel services based on big data provided by an embodiment of this application;

[0045] Figure 4 is a schematic structural diagram of a computer device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0047] To facilitate a clearer understanding of the present application, first, the application scenario of the hotel service push method based on big data for implementing the present application will be introduced. As Figure 1 shown, this application scenario includes a computer device 10 and a client cluster. The client cluster can include one or more clients, and the number of clients will not be limited here. As Figure 1 shown, the terminal cluster can specifically include client 1, client 2,..., client n. It can be understood that client 1, client 2, client 3,..., client n can all be network-connected to the computer device 10, so that each client can perform data interaction with the computer device 10 through the network connection.

[0048] It can be understood that the computer device 10 can refer to a device for performing requirement analysis, and this computer device 10 can also be used to store user behavior data. The computer device can be a server. Among them, the server can be an independent physical server, or a server cluster or distributed system composed of at least two physical servers, 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 network (CDN), and big data and artificial intelligence platforms. The client can specifically refer to in-vehicle terminals, smart phones, tablet computers, laptop computers, desktop computers, smart speakers, screen speakers, smart watches, etc., but is not limited thereto. Each client and the computer device can be directly or indirectly connected through wired or wireless communication methods. At the same time, the number of clients and the computer device can be one or at least two, and the present application does not make any restrictions here.

[0049] Furthermore, please refer to Figure 2 , which is a schematic flowchart of a hotel service push method based on big data provided by an embodiment of the present application. As Figure 2 shown, this method can be executed by the Figure 1 computer device therein. Among them, the hotel service push method based on big data can include the following steps:

[0050] Step S110: Obtain the big data of target user behavior and load the big data of target user behavior into a hotel service demand analysis model that has been debugged in advance. Among them, the big data of target user behavior includes the hotel interaction behavior data of the target user.

[0051] Among them, obtaining the big data of target user behavior means collecting the behavior data of the target user through specific technical means from various sources (such as server logs, user devices, third-party applications, etc.). These data are huge in scale, diverse in types, and extremely fast in generation speed, so they are called "big data". For example, an online travel platform can collect information such as the search history of users on its website, the hotel detail pages clicked, and the hotel reviews browsed to understand the preferences and behavior patterns of users. The big data of target user behavior refers to the behavior data collected for a specific user group or a single user, and these data have clear analysis purposes and pertinence. For example, if a hotel group wants to understand the accommodation preferences of its VIP customers, it can collect data such as the past reservation records, the types of hotels stayed in, and room selections of these VIP customers as the big data of target user behavior.

[0052] The hotel service demand analysis network that has been debugged in advance is a computer device or software framework that has been tested and debugged and is ready to receive input data and perform analysis. It is specifically designed to analyze the user needs related to hotel services. For example, a machine learning-based hotel recommendation system needs to undergo a large amount of data training and algorithm debugging before going online to ensure that it can accurately analyze the hotel service needs of users and give appropriate recommendations. Hotel interaction behavior data refers to the data generated when users interact on online platforms or applications related to hotels. These interaction behaviors may include searching, clicking, booking, evaluating, etc. For example, a user searches for "five-star hotels" on a hotel reservation platform, views the detail pages of several hotels, and finally books a hotel located in the city center. These search, view, and booking behaviors will all be recorded as hotel interaction behavior data.

[0053] The following is a specific example that describes the expression of a user behavior data in a structured data format. Suppose a computer device has a hotel reservation platform that is tracking the behavior of users to optimize service push.

[0054] Construct a data table based on the behavioral data of users booking hotels. Assume the name of the data table is: user_booking_events. The data fields include: event_id: event ID (unique identifier); user_id: user ID (unique identifier); event_type: event type; timestamp: timestamp (the time when the event occurred); hotel_id: hotel ID (the hotel booked or viewed by the user); room_type: room type; price: price; booking_status: booking status (such as "booked", "not booked", "cancelled", etc.).

[0055] Referring to the following table, a specific data record is as follows:

[0056] In this example, user A123 first clicked on the Deluxe room type of hotel H001 (event ID 1), then viewed the details of this room type (event ID 2), and finally booked this room type (event ID 3). User B456 directly clicked and booked the Standard room type of hotel H002 (event IDs 4 and 5). This structured data format enables computer devices to easily query and analyze user behavioral data. For example, the click-to-booking conversion rate for each room type can be calculated, or the most popular hotels and room types within a specific time period can be analyzed. This data is very valuable for personalized service push, price optimization, and market strategy formulation.

[0057] To convert the above-mentioned behavioral data of users booking hotels into a format that can be processed by a machine learning model (i.e., a hotel service demand analysis network), a computer device can first perform operations such as data cleaning, feature engineering, data transformation, data normalization / standardization, etc. In the above example, the data is complete without missing values or outliers, so the computer device can directly perform feature engineering. Useful features may include user ID (used to identify the behavioral patterns of different users), event type (click, view details, book, etc., which can be converted into one-hot encoding), timestamp (features such as hour, day of the week, month, etc. can be extracted), hotel ID (used to identify the popularity of different hotels), room type (which can be converted into one-hot encoding or categorical encoding), price (numerical feature, which may need to be normalized), etc. In this scenario, the computer device can temporarily ignore booking_status because it can be inferred from the event type.

[0058] For data transformation, convert categorical data (such as event type and room type) into numerical data. This can usually be achieved through one-hot encoding or label encoding. In an example, the process of one-hot encoding example (applicable to event type and room type) is as follows:

[0059] For event types (event_type), it can include click, view_details, book, etc., which are encoded as: click > [1, 0, 0], view_details > [0, 1, 0], book > [0, 0, 1]; for room types (room_type), it can be encoded as: Deluxe > [1, 0], Standard > [0, 1].

[0060] In an example, the process of the label encoding example is as follows:

[0061] For event types (event_type), it can include click, view_details, book, etc., which are encoded as: click > 1, view_details > 2, book > 3; for room types (room_type), it can be encoded as: Deluxe > 1, Standard > 2.

[0062] After that, the data can be normalized / standardized. For numerical features such as price, the computer device may need to normalize or standardize it so that all features are on the same scale. For example, the computer device can use min - max normalization to scale the price between 0 and 1, or use z - score normalization. For min - max normalization, the formula can be:

[0063] normalized_price = (price - min_price) / (max_price - min_price)

[0064] For Z - score normalization, the formula can be:

[0065] standardized_price = (price - mean_price) / std_dev_price

[0066] In an example of the conversion result, the obtained result is:

[0067] In this example, assume there are only two room types (Deluxe and Standard), and the price has been normalized between 0 and 1. In fact, hotel IDs may need further processing (e.g., through target encoding or embedding) because they are categorical features and may have a large number of different values. In addition, user IDs may not be used as direct features in the actual model but are used for aggregation or building user - level features.

[0068] Step S120: Analyze the big data of the target user's behavior based on the hotel service demand analysis model to obtain the hotel service demands of the target user.

[0069] Step S130: Based on the hotel service demands of the target user, determine the target hotel service that matches the hotel service demands among the preset hotel services, and push the target hotel service to the client logged in by the target user.

[0070] In this step S130, the computer device, according to the specific needs of the user (hotel service demands), finds the services that match the user's needs (target hotel services) among a group of predefined hotel service options, and then actively sends the information of these matching services to the device or application interface (client) that the user is using.

[0071] The specific requirements and expectations of users for hotel services can include multiple aspects such as price, location, facilities, and services. For example, the hotel service demands of a user can be "book a standard room located in the city center, with a price not exceeding 500 yuan per night, and with free Wi-Fi and breakfast service". The preset hotel services are the hotel service options predefined in the system, and these options are usually set based on the actual situation of the hotel and cooperation agreements. For example, the preset hotel services can include detailed information such as room types, prices, and service facilities of multiple hotels, such as "a luxury double room in a certain hotel, including free breakfast and use of the gym, with a price of 600 yuan per night". The computer device can find the service options that match the specific needs of the user through algorithms or manually. For example, the system filters out several hotels that meet the conditions according to the user's budget, location preferences, etc., and further compares conditions such as facilities and services, and finally determines a hotel that best meets the user's needs as the target hotel service. When the user browses the hotel reservation APP, the system actively pushes hotel discount information or room type recommendations that meet the user's needs to the user. The client logged in by the target user is the software application or device interface that the user is currently logged in and using, and it is the medium for the user to interact with the system. For example, the user opens the hotel reservation APP on their smartphone and successfully logs in to the account, and this APP interface is the client logged in by the target user, and the system will display the pushed target hotel service information to the user through this interface.

[0072] Based on the above steps S110 - S130, the embodiments of the present application provide a solution for analyzing user behavior data using a computer device to provide personalized hotel service recommendations. First, step S110 obtains the big data of target user behavior and loads it into a hotel service demand analysis model that has been pre - debugged. The big data of target user behavior here mainly refers to the data generated by the target user in various interactions related to the hotel, such as the browsing records, reservation records, evaluation records, etc. on the hotel official website. These data can truly reflect the user's preferences and demands for hotel services.

[0073] The hotel service demand analysis model is a machine - learning - based model, which may be a model of types such as decision tree, random forest, neural network, etc. This model has been trained and debugged in advance with a large amount of user behavior data, enabling it to have the ability to analyze user behavior data and predict the user's hotel service demands.

[0074] Next, step S120 analyzes the big data of target user behavior based on the hotel service demand analysis model to obtain the hotel service demands of the target user. By calculating and processing the user data through the model, the user's behavior patterns and service demands are extracted. Finally, step S130 determines the target hotel service that matches the hotel service demands among the preset hotel services according to the hotel service demands of the target user, and pushes the target hotel service to the client logged in by the target user. For example, if the user's demand is a luxury suite, then the system will push the information of the luxury suite that meets the demand to the user.

[0075] Among them, the hotel service demand analysis model is determined based on the behavior data template library. The behavior data template library includes a positive - type behavior data template library and a to - be - classified - type behavior data template library. The hotel service demand analysis model is obtained through the following steps of debugging:

[0076] Step S10: Configure the prior labels of the behavior data templates of each positive type as positive types, configure the prior labels of the behavior data templates of each type to be classified as negative types, and based on the prior labels of each behavior data template, use the behavior data template library as a training batch to iteratively debug the hotel service demand analysis model, obtaining a first demand analysis model and the demand analysis values inferred by the first demand analysis model for the behavior data templates of each type to be classified, as well as a second demand analysis model and the demand analysis values inferred by the second demand analysis model for the behavior data templates of each type to be classified, where the first demand analysis model is obtained by using the first number of iterative debuggings, the second demand analysis model is obtained by using the second number of iterative debuggings, the first number is not equal to the second number, and in the iterative debugging, optimize the model parameter variables of the hotel service demand analysis model based on the demand determination cost of the behavior data template library.

[0077] Among them, the behavior data template library is a set containing multiple behavior data templates, and the behavior data template is a training sample used in the model debugging process. Each behavior data template contains specific behavior information of the user's interaction with the hotel, such as browsing history, booking records, evaluation feedback, etc., and this information has been formatted into a format that the model can understand and process. Its format can refer to the example of the target user behavior big data in step S110. During the training process, the model will learn to extract useful features from these behavior data templates and make predictions or classifications based on these features. In the behavior data template library, the positive type refers to those behaviors that clearly show the user's preferences or needs, such as frequently booking the same type of hotel; while the type to be classified refers to behaviors with unclear behavior patterns that require further analysis to determine the user's needs. The prior label refers to the pre-classification label of the behavior data template based on known information (such as business logic, historical data, etc.) before providing training data for the model. In this example, positive and negative are two possible values of the prior label. The demand analysis value is a quantitative indicator output by the model, which reflects the model's inference of the hidden needs behind specific user behaviors. This value may be a probability, a score, or other forms of quantitative representation. The model parameter variables are the variables inside the model, and their values are optimized during the training process to minimize prediction errors. The parameter variables can include weights, biases, learning rates, etc., and they jointly determine how the model interprets the input data and makes predictions. The first number and the second number refer to different numbers of iterative debuggings selected during the iterative debugging process. By comparing the performance of the model at different numbers of iterative debuggings, the stability and overfitting risk of the model can be evaluated.

[0078] When the computer device executes step S10, it first processes the behavioral data template library. This library contains user behavioral data collected from multiple sources such as hotel reservation systems and user review platforms. After being cleaned and formatted, these data form standardized behavioral data templates. Next, the computer device marks these templates based on prior knowledge. For example, if a user has booked high-end hotels multiple times and left positive reviews in the past year, their behavioral data template may be marked as "positive". On the contrary, if a user only browses the hotel page occasionally without further interaction, their template may be marked as "negative".

[0079] Then, the computer device starts an iterative debugging process. In this process, the hotel service demand analysis model (which may be a deep learning model, such as a recurrent neural network RNN, suitable for processing sequential data like user behavioral history) uses these marked templates as input for training. The model attempts to learn the mapping relationship from behavioral data to demand labels. After each iteration, the model evaluates its performance and adjusts its parameters according to the defined cost function. This cost function may be cross-entropy loss, which is used to measure the difference between the probability distribution predicted by the model and the true labels.

[0080] By comparing the performance of the model at different iteration numbers (such as the first number and the second number), the computer device can evaluate whether the model is overfitting or underfitting and adjust the training strategy accordingly. Eventually, after multiple iterations and optimizations, the model will be able to more accurately infer users' hotel service demands from their behavioral data.

[0081] By implementing the above step S10, the hotel service demand analysis model can finely capture the nuances of user behavior and convert them into clear demands for hotel services. This not only enhances the model's prediction ability but also provides a powerful tool for hotels to better understand and serve their customers. In addition, this solution ensures that the model remains highly flexible and accurate in the face of complex and changing user behaviors by combining prior knowledge and iterative learning.

[0082] In other words, generally speaking, for each type of behavior data template in the behavior data template library, prior markings are configured according to its nature. Among them, the behavior data templates of the positive type are marked as the positive type, while the behavior data templates of the type to be classified are marked as the negative type. These prior markings will serve as the initial reference for model training. Next, the behavior data template library with the configured prior markings is used as a training batch to iteratively debug the hotel service demand analysis model. In this process, the model will continuously learn and adjust to better adapt to the training data. Through iterative debugging, two different versions of the demand analysis model can be obtained: the first demand analysis model and the second demand analysis model. These two models are obtained at different numbers of iterations, so their parameters and performances will also be different. At the same time, these two models will also infer the corresponding demand analysis values for each behavior data template of the type to be classified. During the iterative debugging process, it is also necessary to optimize the model parameters based on the demand determination cost of the behavior data template library. Here, the "demand determination cost" is the error or loss of the model when inferring the demand analysis value. By minimizing this cost, the model can more accurately predict the user's needs.

[0083] For example, to illustrate this process: Suppose the behavior data template library contains data templates for behaviors such as users browsing hotel pages, booking hotels, and evaluating hotels. In step S10, the model will first learn these behavior data templates and adjust its parameters according to the prior markings (positive or negative). Then, the model will try to infer the user needs (such as the user may be interested in the hotel but still hesitating) behind each behavior data template of the type to be classified (such as the user browsed the page of a certain hotel but did not book). By continuously iteratively debugging and optimizing the parameters, the model can gradually improve its inference accuracy.

[0084] As an implementation method, among them, according to the prior markings of each behavior data template, using the behavior data template library as a training batch to iteratively debug the hotel service demand analysis model, obtaining the first demand analysis model and the demand analysis values inferred by the first demand analysis model for each behavior data template of the type to be classified, as well as the second demand analysis model and the demand analysis values inferred by the second demand analysis model for each behavior data template of the type to be classified, includes the following steps:

[0085] Step S11a: In the first - numbered iterative debugging of the hotel service demand analysis model, determine the demand determination value of the last iterative debugging for each behavior data template of the type to be classified as the demand analysis value inferred by the first demand analysis model for each behavior data template of the type to be classified;

[0086] Step S12a: In the second iterative debugging of the hotel service demand analysis model, determine the demand determination value of the behavior data template for each type to be classified in the last iterative debugging as the demand analysis value inferred by the second demand analysis model for the behavior data template of each type to be classified.

[0087] Through the execution processes of steps S11a and S12a, the demand determination value of the last time is usually relatively accurate. Taking it as the demand analysis value, the acquisition process is relatively convenient.

[0088] As another implementation manner, wherein, according to the prior label of each behavior data template, use the behavior data template library as a training batch to perform iterative debugging on the hotel service demand analysis model, and obtain the first demand analysis model and the demand analysis value inferred by the first demand analysis model for the behavior data template of each type to be classified, as well as the second demand analysis model and the demand analysis value inferred by the second demand analysis model for the behavior data template of each type to be classified, including:

[0089] Step S11b: According to the pre-deployed eccentricity adjustment coefficient, determine the weighted summation result of the respective demand determination values output by the hotel service demand analysis model for the behavior data template of each type to be classified in the first iterative debugging as the demand analysis value inferred by the first demand analysis model for the behavior data template of each type to be classified;

[0090] Step S12b: According to the pre-deployed eccentricity adjustment coefficient, determine the weighted summation result of the respective demand determination values output by the hotel service demand analysis model for the behavior data template of each type to be classified in the second iterative debugging as the demand analysis value inferred by the second demand analysis model for the behavior data template of each type to be classified.

[0091] Because the prior label of the behavior data template of the type to be classified is set as the negative type, and the behavior data templates in the behavior data template library have noise, making the demand analysis value inferred by the hotel service demand analysis model have reference value during each iterative debugging. The demand analysis value obtained from the last iterative debugging may not necessarily be the most referenceable. Based on this, perform eccentricity adjustment (that is, perform weighted calculation) on each demand classification value, and the accuracy of using the result after eccentricity adjustment (that is, the weighted summation result) as the demand analysis value is higher.

[0092] Alternatively, in another embodiment, based on the prior tags of each behavior data template, the hotel service demand analysis model is iteratively debugged with the behavior data template library as a training batch to obtain a first demand analysis model and the demand analysis values inferred by the first demand analysis model for the behavior data templates of each type to be classified, and a second demand analysis model and the demand analysis values inferred by the second demand analysis model for the behavior data templates of each type to be classified, including:

[0093] For each behavior data template of each type to be classified, the following operations are performed:

[0094] Step S11c: Use the product of the demand determination value of the hotel service demand analysis model for the behavior data template of the type to be classified in the first iterative debugging and a second preset model parameter variable as the demand analysis value inferred by the hotel service demand analysis model for the behavior data template of the type to be classified in the first iterative debugging. The second preset model parameter variable is between 0 and 1.

[0095] Okay, next I will explain Step S11c in detail.

[0096] The first iterative debugging refers to the first complete iterative debugging in the model training process. During this process, the hotel service demand analysis model analyzes each template in the behavior data template library and outputs a demand determination value for that template. This demand determination value is the demand result initially inferred by the model based on the prior tags and other feature information of the template. The second preset model parameter variable is a value between 0 and 1, which is used to weight the demand determination value output by the model in the first iterative debugging. By multiplying with this parameter variable, the demand analysis value of the behavior data template of the type to be classified in the first iterative debugging can be obtained.

[0097] For example, assume that the demand determination value of a certain behavior data template of a type to be classified in the first iterative debugging is 0.8, and the value of the second preset model parameter variable is 0.6. According to the operation of Step S11c, multiplying these two values gives the demand analysis value of the behavior data template of the type to be classified in the first iterative debugging as 0.48 (i.e., 0.8 × 0.6). This processing method helps to introduce a certain degree of conservatism in the early stage of model training and prevent the model from being too confident in the initial inference results. By adjusting the value of the second preset model parameter variable, the sensitivity of the model to the demand determination value in the first iterative debugging can be controlled, thereby affecting the calculation of the final demand analysis value. It should be noted that Step S11c is only applicable to the operations in the first iterative debugging. In subsequent iterative debuggings, other methods (such as described in Step S12c) will be used to calculate the demand analysis values of the behavior data templates of each type to be classified.

[0098] Step S12c: Determine the requirement analysis value of the behavior data template of the to-be-classified type in other sub-iteration debugging inferences of the hotel service requirement analysis model based on the following formula:

[0099] h n h(i) = a·h m (i) + b·f n (i),

[0100] where i represents the behavior data template of the to-be-classified type, n represents the round number of iterative debugging, h n h(i) is the requirement analysis value of the behavior data template of the to-be-classified type in the nth iterative debugging inference of the hotel service requirement analysis model, h m h(i) is the requirement analysis value of the behavior data template of the to-be-classified type in the mth iterative debugging inference of the hotel service requirement analysis model, a and b are the second preset model parameters, f n f(i) is the requirement determination value of the behavior data template of the to-be-classified type in the nth iterative debugging of the hotel service requirement analysis model; where b = 1 - a, m = n - 1.

[0101] The core idea of this formula is to combine the requirement analysis value of the previous iteration and the requirement determination value of the current iteration, and obtain the requirement analysis value of the current iteration through weighted summation. The advantage of doing this is that the model can consider both historical information and current information at the same time, making the requirement analysis value smoother and more stable during the iterative process. For example, assume that in a certain iterative round n, for a behavior data template i of a to-be-classified type, the requirement analysis value h m h(i) of the previous iteration is 0.7, and the requirement determination value f n f(i) of the current iteration is 0.8, and the second preset model parameters a and b are set to 0.6 and 0.4 respectively (because b = 1 - a). According to the formula in Step S12c, the requirement analysis value h n h(i) of this template in the current iteration can be calculated as:

[0102] h n h(i) = 0.6×0.7 + 0.4×0.8 = 0.42 + 0.32 = 0.74

[0103] This means that, after considering the requirement analysis value of the previous iteration and the requirement determination value of the current iteration, the model believes that the requirement analysis value of this template in the current iteration is 0.74. This process will be repeated in each iteration, continuously updating the requirement analysis value of each template until the model converges or reaches the preset number of iterations. It should be noted that step S12c applies to all iteration processes except the first iteration. In the first iteration, since there is no requirement analysis value of the previous iteration for reference, the method in step S11c will be used to calculate the requirement analysis value.

[0104] Step S13c: Use the requirement analysis value of the behavior data template of this type to be classified in the first number of iterations of debugging of this hotel service requirement analysis model as the requirement analysis value inferred by this first requirement analysis model for the behavior data template of this type to be classified.

[0105] Step S13c focuses on formally establishing the requirement analysis value calculated in a specific number of iterations (here referring to the first number, i.e., the first iteration) as the inference output of the first requirement analysis model for a certain type of behavior data template to be classified. Specifically, when the hotel service requirement analysis model conducts the first iteration of debugging, it will calculate the requirement analysis value for each type of behavior data template to be classified. This calculation process will comprehensively consider the prior markings of the template, model parameters, and other factors that may affect requirement inference. After the calculation is completed, the model will output a preliminary requirement analysis value for each behavior data template.

[0106] The task of step S13c is to fix the requirement analysis value of each behavior data template of the type to be classified obtained in this iteration as the final inference result of the first requirement analysis model for this type of template. This means that in subsequent model applications or further analyses, if the first requirement analysis model is used to infer the same type of behavior data template, it will directly output the requirement analysis value determined in the first iteration. For example, assume that in the first iteration of debugging, the requirement analysis value calculated by the model for the behavior data template of the type to be classified "booking room type" is 0.85 (this value may represent a certain requirement intensity or probability). Then according to step S13c, this 0.85 is established as the inference output of the first requirement analysis model for all "booking room type" templates.

[0107] Such a processing method helps to establish a stable and consistent requirements analysis baseline, especially in the early stage of model training. As the iteration progresses and the model is optimized, subsequent requirements analysis models may revise or refine these preliminary reasoning results, but the values established in the first iteration still have important reference value. It should be noted that the first requirements analysis model here does not specifically refer to a certain specific machine learning model or neural network structure, but refers to a model snapshot or version established based on the data of the first iteration during the entire model construction and debugging process.

[0108] Step S14c: Use the requirements analysis value of the behavior data template of the type to be classified in the second iteration debugging of the hotel service requirements analysis model as the requirements analysis value inferred by the second requirements analysis model for the behavior data template of the type to be classified.

[0109] The core objective of step S14c is to establish the requirements analysis value obtained in the second iteration debugging as the inference output of the second requirements analysis model for a specific behavior data template of a certain type to be classified. Specifically, when the hotel service requirements analysis model reaches the second iteration debugging, it will calculate and analyze each type of behavior data template to be classified again to update and optimize the requirements analysis of the model for these templates. In this process, the model will use the experience and information obtained from the first iteration, as well as possible new data or adjusted parameters, to re-evaluate the requirements of each type of template. After completing the second iteration debugging, the model will generate a new requirements analysis value for each behavior data template. The task of step S14c is to fix the requirements analysis value of each behavior data template of the type to be classified calculated in this iteration as the final inference result of the second requirements analysis model for this type of template. This means that in subsequent model applications or further analyses, if the second requirements analysis model is used to infer the same type of behavior data template, it will directly output the requirements analysis value determined in the second iteration.

[0110] For example, assume that in the second iteration of debugging, the demand analysis value calculated by the model for the behavior data template of the to-be-classified type "reserving breakfast" is 0.9 (this value may represent a certain demand intensity or probability). Then, according to step S14c, this 0.9 is established as the inference output of the second demand analysis model for all "reserving breakfast" type templates. Such a processing method helps the model to gradually optimize its demand analysis ability during continuous iteration, while maintaining the consistency and comparability of the inference results at each iteration stage. As the number of iterations increases, the model is expected to gradually approach the true demand distribution, improving the accuracy and effectiveness of the analysis. It should be emphasized that the second demand analysis model mentioned here does not refer to a specific machine learning model or neural network structure, but rather a model snapshot or version established based on the data of the second iteration during the entire model construction and debugging process.

[0111] Step S20: Determine the fuzzy index of each behavior data template of the to-be-classified type based on the error between the demand analysis values inferred by the first demand analysis model and the second demand analysis model for each behavior data template of the to-be-classified type in the behavior data template library.

[0112] Among them, the demand analysis value refers to the result output by the model after inferring the user behavior data template, which reflects the model's judgment or prediction of the user's needs. This value can be continuous (such as a probability score) or discrete (such as a classification label). For example, the first demand analysis model and the second demand analysis model may respectively output a demand analysis value indicating the degree of the user's demand for a high-end hotel or an economy hotel. For example, a demand analysis value may be 0.8, indicating that the model has 80% confidence that the user prefers a high-end hotel. The error refers to the difference between the demand analysis values generated when two or more models infer the same data point. The error can quantify the inconsistency between models, thus reflecting the uncertainty of the models' interpretation of the data. For example, if the demand analysis value given by the first demand analysis model is 0.8 (preference for high-end hotels), and the second demand analysis model gives 0.3 (preference for economy hotels), then the error between the two is 0.5. This error value indicates that there is a large disagreement between the two models in predicting the user's needs.

[0113] The Fuzziness Index is a quantitative metric that represents the uncertainty or ambiguity of a model's classification of a specific data point. It is typically calculated based on the error between different models or other uncertainty measures. For example, the Fuzziness Index can be calculated based on the error between the first requirements analysis model and the second requirements analysis model. If the error is large, the Fuzziness Index will be correspondingly high, indicating that the model's judgment of the user's requirements is relatively fuzzy or uncertain. For example, a Fuzziness Index of 0.7 may mean that there is a significant uncertainty in whether the model determines that the user prefers a high-end hotel or an economy hotel. The core task of step S20 is to determine the Fuzziness Index of the behavior data template for each type to be classified. This process is based on two previously created requirements analysis models: the first requirements analysis model and the second requirements analysis model. These two models may be machine learning models, such as decision trees, random forests, support vector machines, or neural networks such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), and the specific choice depends on the characteristics of the data and the complexity of the problem. However, for the sake of simplicity in explanation, specific model details are not involved, but rather the logic of step S20 is focused on. In step S20, the first requirements analysis model and the second requirements analysis model are used to perform inferences on each template in the behavior data template library for the type to be classified. Here, "inference" means that the model processes the input data (i.e., the behavior data template) based on the learned knowledge and patterns and outputs a requirements analysis value. This value reflects the model's judgment of the user requirement type to which the template belongs. Since there are two different requirements analysis models, they may produce different requirements analysis values for each behavior data template. An important part of step S20 is to compare these two values and determine the difference or error between them. This error may stem from the different structures of the models, differences in the training data, or the complexity and ambiguity of user behavior.

[0114] Finally, step S20 quantifies this error into a Fuzziness Index. The Fuzziness Index is a numerical value that represents the uncertainty of the model regarding the requirement type to which a specific behavior data template belongs. The higher the Fuzziness Index, the greater the difference in the requirements analysis of the template by the two models, and thus the lower the confidence of the model in classifying it. Conversely, if the Fuzziness Index is low, it indicates that the requirements analysis of the template by the two models is relatively consistent, and the model has a higher confidence in classifying it.

[0115] For example, assume that a behavioral data template describes that a user frequently browsed the pages of high-end hotels in the past week and finally booked a five-star hotel. The first demand analysis model may classify this template as a "high-end hotel preference person" and give a relatively high demand analysis value; while the second demand analysis model, due to some reasons (such as the deviation in model training), classifies it as a "frequent visitor to economy hotels" and gives a relatively low demand analysis value. In this case, step S20 will calculate the error between these two values and obtain a relatively high fuzzy index, indicating the uncertainty of the model about the demand type to which this specific user behavior template belongs. Based on this, step S20 not only provides the classification result of a single behavioral data template, but also quantifies the uncertainty of the classification, providing more comprehensive information for subsequent decision-making.

[0116] Step S30: Adjust the prior label of each behavioral data template of the to-be-classified type based on the demand analysis value inferred by the first demand analysis model for the behavioral data template of each to-be-classified type.

[0117] As mentioned above, the prior label is a label or classification pre-assigned to each behavioral data template. This label may be set based on some preliminary analyses or assumptions, and it represents a preliminary understanding of user needs. However, the prior label is not necessarily completely accurate, and there may be certain deviations or ambiguities. Therefore, the goal of step S30 is to correct or adjust these prior labels based on the inference results of the first demand analysis model. Specifically, the first demand analysis model will infer each behavioral data template of the to-be-classified type and output a demand analysis value. This demand analysis value reflects the model's judgment or prediction of the user needs to which the template belongs. Then, this demand analysis value is compared with the prior label of the template, and the prior label is adjusted according to the consistency or difference between them. If the demand analysis value is highly consistent with the prior label, then the confidence of the prior label can be retained or enhanced. On the contrary, if there is a large difference between the demand analysis value and the prior label, then we may weaken or modify the prior label to make it closer to the result of the model inference. In a detailed example, for a certain behavioral data template, its prior label is "frequent visitor to economy hotels". Then, the first demand analysis model infers this template and the output demand analysis value indicates that the user actually has a greater preference for booking high-end hotels. In this case, step S30 will adjust the prior label of this template to make it closer to "high-end hotel preference person".

[0118] Step S40: Based on the fuzzy index and the adjusted prior label of each behavioral data template of the to-be-classified type, use the behavioral data template library of this to-be-classified type as a training batch to iteratively debug the first demand analysis model to obtain the debugged first demand analysis model.

[0119] Specifically, during the model training process, the fuzzy index and the adjusted prior label calculated in the previous steps S20 and S30 are used to optimize the parameters of the model. Specifically, the fuzzy index can be used to adjust the weight update of the model during training, enabling the model to pay more attention to those data templates with higher classification uncertainty. The adjusted prior label, on the other hand, serves as the target output to guide the model to learn more accurate classification boundaries.

[0120] Through multiple iterative trainings, the first demand analysis model will gradually adjust its internal parameters to better fit the training data and improve the classification performance. For example, optimization algorithms such as gradient descent can be used to minimize the prediction errors of the model. Eventually, after sufficient iterative debugging (such as the set maximum number of iterations), the debugged first demand analysis model is obtained. The first demand analysis model can not only better handle the previous classification ambiguity problems but also more accurately predict and interpret the user's demand behavior.

[0121] Step S50: Determine the model parameters of the hotel service demand analysis model based on the model parameters of the debugged first demand analysis model.

[0122] Specifically, based on the aforementioned steps S10 - S40, the first demand analysis model has been debugged and optimized to better understand and predict the user's behavior and needs. At this time, the model parameters of the hotel service demand analysis model can be determined based on the debugged first demand analysis model. The model parameters mentioned here refer to the parameters and variables used to define the model structure and behavior in a machine learning model. These parameters are optimized during the model training process to minimize the prediction errors of the model and improve the model's performance. Specifically, the parameters in the debugged first demand analysis model can be directly used as the parameters of the hotel service demand analysis model, or these parameters can be used as the initialization parameters of the hotel service demand analysis model and further training and optimization can be carried out. For example, if the first demand analysis model is a deep learning neural network, its model parameters may include the weights, biases, activation functions, etc. of the neural network, and these parameters can be directly applied to the neural network structure of the hotel service demand analysis model, enabling the hotel service demand analysis model to inherit the learning ability and classification performance of the first demand analysis model.

[0123] Alternatively, step S50, determining the model parameters of the hotel service demand analysis model based on the model parameters of the debugged first demand analysis model, includes:

[0124] Step S51: Adjust the prior label of each behavior data template of the to-be-classified type based on the requirement analysis value inferred for the behavior data template of each to-be-classified type by the second requirement analysis model.

[0125] This step can refer to Step S30, except that the object is replaced by the second requirement analysis model.

[0126] Step S52: Based on the fuzzy index and the adjusted prior label of each behavior data template of the to-be-classified type, use the behavior data template library of the to-be-classified type as a training batch to iteratively debug the second requirement analysis model, and obtain the debugged second requirement analysis model.

[0127] Specifically, the fuzzy index of each behavior data template of the to-be-classified type will be considered first. The fuzzy index reflects the degree of certainty of the model's requirement analysis for this type of template. The higher the index, the more ambiguous the model's requirement analysis for this type of template, and more debugging and optimization are needed. At the same time, the adjusted prior label provides more accurate initial classification information for the model, which helps the model better understand and classify behavior data templates. Next, use the behavior data template library of the to-be-classified type as a training batch to iteratively debug the second requirement analysis model. This means that the model will perform multiple trainings and learnings on this specific training set, continuously adjusting its internal parameters and structure to better fit and classify these behavior data templates. During the debugging process, the model may adopt various machine learning algorithms and techniques, such as the gradient descent algorithm, the backpropagation algorithm, etc., to minimize classification errors and improve classification accuracy. In addition, to prevent overfitting and improve the generalization ability of the model, regularization techniques, dropout techniques, etc. may also be adopted. Finally, after multiple iterative debuggings, a debugged second requirement analysis model is obtained. The second requirement analysis model not only has better performance on the training set, but also can provide more accurate and reliable requirement analysis results when facing new and unknown behavior data templates.

[0128] Step S53: For each model parameter variable of the debugged first requirement analysis model, determine the average result between the model parameter variable and the corresponding model parameter variable of the second requirement analysis model as the corresponding model parameter variable of the hotel service requirement analysis model.

[0129] Step S53 combines the advantages of two models (the first demand analysis model and the second demand analysis model) to obtain a more accurate and comprehensive hotel service demand analysis model. Specifically, for each model parameter variable in the debugged first demand analysis model, the corresponding model parameter variable in the second demand analysis model will be found. Then, the average result between these two corresponding model parameter variables is calculated. This average result is not a simple arithmetic mean, but a weighted average or other forms of average calculation based on factors such as the performance of the two models in their respective training processes and weight allocation.

[0130] Through such a processing method, step S53 can effectively integrate the information of the two models and avoid the biases or deficiencies that may exist in a single model. At the same time, since the average calculation can usually smooth the noise and outliers in the data, this step also helps to improve the stability and robustness of the final hotel service demand analysis model.

[0131] For example, assume that there is a model parameter variable in the debugged first demand analysis model representing "the degree of customer demand for room cleanliness", and there is also a corresponding model parameter variable in the second demand analysis model. In step S53, the average result of these two model parameter variables will be calculated as the corresponding model parameter variable in the hotel service demand analysis model. In this way, when the final hotel service demand analysis model evaluates the customer's demand for room cleanliness, it can consider the information of both models at the same time, thus obtaining a more accurate and comprehensive conclusion.

[0132] In some embodiments, step S30, adjusting the prior label of each behavior data template of each type to be classified based on the demand analysis value inferred by the first demand analysis model for the behavior data template of each type to be classified, includes:

[0133] For each behavior data template of each type to be classified in the behavior data template library of the type to be classified, when the demand analysis value inferred by the first demand analysis model for the behavior data template of the type to be classified is greater than the positive type threshold, the prior label of the behavior data template of the type to be classified is adjusted to the positive type; when the demand analysis value inferred by the first demand analysis model for the behavior data template of the type to be classified is not greater than the positive type threshold, the prior label of the behavior data template of the type to be classified is adjusted to the negative type.

[0134] Specifically, when the first requirement analysis model infers a behavior data template of a certain type to be classified and the obtained requirement analysis value is greater than the preset positive type threshold value, this means that the user behavior or requirement represented by this template is highly correlated with the positive type. Therefore, the prior label of this type of template is adjusted to the positive type to reflect its actual attributes. On the contrary, if the requirement analysis value inferred by the first requirement analysis model is not greater than the positive type threshold value, this indicates that the correlation between this template and the positive type is low, or it is more likely to belong to the negative type. In this case, the prior labels of these templates are adjusted to the negative type.

[0135] For example, assume a behavior data template library of a certain type to be classified, which contains multiple user behavior data templates related to hotel services. Among them, a certain template represents the behavior of a user frequently booking high-star hotel suites. If the requirement analysis value inferred by the first requirement analysis model for this template is very high, far exceeding the positive type threshold value, then the prior label of this template is adjusted from the original type to be classified to the positive type, because this behavior is usually associated with high-end and high-satisfaction hotel service requirements. In this way, not only the prior labels of the behavior data templates of each type to be classified are optimized, but also the sensitivity and response speed of the model to actual user requirements are improved. In addition, this dynamic label adjustment strategy based on the model inference results also helps to enhance the adaptability and accuracy of the entire hotel service requirement analysis model, enabling it to better serve the changing user requirements and market environment.

[0136] Based on this, as an implementation method, step S40, according to the fuzzy index and the adjusted prior label of the behavior data template of each type to be classified, using the behavior data template library of this type to be classified as a training batch to perform iterative debugging on the first requirement analysis model, may include:

[0137] Step S41: Determine the importance coefficient of the behavior data template of each type to be classified based on the fuzzy index of the behavior data template of each type to be classified, and this importance coefficient is opposite to the change trend of the fuzzy index.

[0138] Specifically, the importance coefficient can be understood as the influence weight of different data templates on the optimization of model parameters during the model training process. The fuzzy index is a quantitative indicator that reflects the uncertainty of the model in classifying a certain data template. Specifically, if the fuzzy index of a behavioral data template of a type to be classified is relatively high, it means that the model is less certain about the classification result of this template, and there is a greater risk of error. Therefore, in step S41, a lower importance coefficient will be assigned to such templates to reduce their influence weight in the subsequent model parameter optimization. On the contrary, for those data templates with a lower fuzzy index (i.e., the model is more certain about their classification results), higher importance coefficients will be assigned to them to enhance their influence in the model parameter optimization. For example, assume a behavioral data template library of a type to be classified, which contains three templates: A, B, and C. After the previous steps, the calculated fuzzy indices of them are 0.9, 0.5, and 0.1 respectively. In step S41, an importance coefficient is assigned to each template according to these fuzzy indices. Assume a simple linear mapping method is adopted to map the fuzzy index to the importance coefficient in the interval [0,1], where the higher the fuzzy index, the lower the importance coefficient. Then, the importance coefficients of templates A, B, and C may be set to 0.1, 0.5, and 0.9 respectively.

[0139] In this way, step S41 provides crucial weight guidance for the subsequent model parameter optimization process, ensuring that those data templates with more certain classification results play a greater role in the model training, while those templates with greater classification uncertainty are suppressed to a certain extent, thereby preventing them from causing excessive interference to the model. This strategy helps to improve the training efficiency and classification accuracy of the model.

[0140] In one embodiment, the fuzzy index of each behavioral data template of each type to be classified is not less than zero. Step S41 determining the importance coefficient of each behavioral data template of each type to be classified based on the fuzzy index of each behavioral data template of each type to be classified may include: taking the ratio of the opposite number of the fuzzy index of each behavioral data template of each type to be classified to the first preset model parameter variable as the exponent of a preset value to determine the importance coefficient of each behavioral data template of each type to be classified.

[0141] For each behavior data template of the type to be classified, there is a corresponding fuzzy index, and its calculation method will be introduced in the subsequent process. The higher the fuzzy index, the more uncertain the model is about classifying this template. Next, the ratio of the negative value of the fuzzy index of each behavior data template of the type to be classified to the first preset model parameter variable is taken as the exponent of the preset value. Here, the first preset model parameter variable is a preset constant used to adjust the relationship between the fuzzy index and the importance coefficient. The preset value is a constant used as the base of the exponential function, usually a number greater than 1, such as the natural constant.

[0142] The calculation formula of the importance coefficient can be expressed as:

[0143]

[0144] Among them, the negative value of the fuzzy index is obtained by taking the negative of the fuzzy index. The purpose of doing this is to make the fuzzy index and the importance coefficient show an opposite trend of change, that is, the higher the fuzzy index (the greater the classification uncertainty), the lower the importance coefficient.

[0145] The value calculated by the above formula is the importance coefficient of each behavior data template of the type to be classified. This importance coefficient will play a role in the subsequent model parameter optimization and is used to adjust the weights of different data templates during training. For example, assume there is a behavior data template of the type to be classified with a fuzzy index of 0.8, the first preset model parameter variable is set to 0.5, and the preset value is set to 2. According to the formula calculation, the importance coefficient of this template is:

[0146]

[0147] This means that in the subsequent model debugging, the weight of this template is relatively low because the model has a large degree of uncertainty in its classification result. In this way, it can help to more reasonably allocate the weights of different data templates in model training, thereby improving the training efficiency and classification accuracy of the model.

[0148] Then, the following operations are repeatedly performed until the first requirement analysis model reaches the debugging cut-off condition:

[0149] Step S42: Determine the requirement determination value of each behavior data template of the type to be classified based on the first requirement analysis model.

[0150] Specifically, in step S42, the first demand analysis model generates a demand determination value for each behavior data template of each type to be classified. This demand determination value is a predicted numerical value, which represents the quantitative judgment made by the first demand analysis model on the user demand represented by the behavior data template according to the current parameters and training data. Specifically, the demand determination value refers to a specific numerical value output by the model when attempting to understand and classify user behavior data. This numerical value is based on the model's analysis of the current data, as well as its learning and experience from past similar data. In a sense, this demand determination value can be regarded as the model's guess or prediction of the user demand.

[0151] In step S42, the model outputs a demand determination value for each such behavior data template. This value may be a numerical value between 0 and 1, where 0 indicates that the user has no intention at all, and 1 indicates that the user has a very strong intention. Of course, the actual numerical range and interpretation may vary depending on the model and application scenario. Take a specific example. Suppose there is a behavior data template that represents a user who has searched for a certain hotel service three times in the past week and browsed the relevant service pages. In step S42, the first demand analysis model outputs a demand determination value for this template, such as 0.7. This numerical value represents the model's prediction of the user's willingness to purchase the service in the future.

[0152] Step S43: Determine the demand determination cost of each behavior data template of each type to be classified based on the prior label and demand determination value of each behavior data template of each type to be classified.

[0153] Specifically, the demand determination cost is an indicator that measures the gap between the model's prediction result and the true result. In step S43, first, the prior label of each behavior data template needs to be obtained. This prior label is usually manually annotated or obtained from other reliable sources, and it represents the true situation of the user demand corresponding to this behavior data template. Then, the demand determination value generated by the model in step S42 is compared with this prior label, and the gap between them is calculated. This gap is the demand determination cost.

[0154] The calculation formula for the demand determination cost can be expressed as:

[0155] C = f(L, P)

[0156] where C represents the demand determination cost, L represents the prior label of the behavior data template, P represents the demand determination value generated by the model, and f is a cost function used to calculate the gap between L and P.

[0157] The specific form of the cost function f can be designed according to the actual application scenarios and requirements. A common approach is to define the requirement determination cost as the absolute error or squared error between the prior label and the requirement determination value, i.e.:

[0158] C = |L - P| or C = (L - P) 2

[0159] It should be noted that both the prior label and the requirement determination value usually need to be normalized to ensure that they are compared on the same scale.

[0160] For example, assume a behavior data template with a prior label of 0.8 (indicating a high purchase intention of the user) and a requirement determination value of 0.6 generated by the model. If the absolute error is used as the cost function, then the requirement determination cost of this template is:

[0161] C = |0.8 - 0.6| = 0.2

[0162] This cost value represents the error degree of the model in predicting the user's needs. In the subsequent model optimization process, the model parameters will be adjusted according to this cost value to reduce the prediction error and improve the accuracy of the model.

[0163] Through step S43, the requirement determination cost of each behavior data template of the type to be classified can be obtained, and these cost values will be used in the subsequent steps to further optimize the model parameters and improve the classification performance of the model.

[0164] Step S44: Determine the requirement determination cost of the behavior data template library of the type to be classified according to the requirement determination cost and importance coefficient of each behavior data template of the type to be classified.

[0165] Step S44 is to evaluate the accuracy of the model's requirement analysis for this type of behavior data as a whole and provide guidance for subsequent model optimization. In step S44, first, the requirement determination cost calculated in step S43 for each behavior data template of the type to be classified, and the importance coefficient determined in step S41 need to be obtained. These two values represent the error degree of the model's prediction result for this template and the importance degree of this template in model training respectively. Next, multiply the requirement determination cost of each behavior data template by its importance coefficient to obtain the weighted requirement determination cost. This weighting process is to ensure that those templates that are more important for model training have a greater weight in the overall cost calculation. Finally, add up all the weighted requirement determination costs to obtain the requirement determination cost of the entire behavior data template library of the type to be classified. This total cost value represents the accuracy of the model's overall requirement analysis for this type of behavior data.

[0166] The calculation formula for determining the cost of requirements can be expressed as:

[0167]

[0168] Among them, T represents the requirement determination cost of the behavior data template library of the type to be classified, n represents the number of behavior data templates of this type, Ci represents the requirement determination cost of the i-th behavior data template, and Wi represents the importance coefficient of the i-th behavior data template.

[0169] For example, assume there is a behavior data template library of the type to be classified, which contains three templates: A, B, and C. In step S43, the calculated requirement determination costs are 0.2, 0.5, and 0.3 respectively; in step S41, the determined importance coefficients are 0.5, 0.3, and 0.2 respectively. Then, according to the calculation formula in step S44, the requirement determination cost of this behavior data template library is:

[0170] T = 0.2×0.5 + 0.5×0.3 + 0.3×0.2 = 0.21 + 0.15 + 0.06 = 0.42

[0171] This total cost value represents the error degree of the model's overall requirement analysis for this type of behavior data, and it will be used as an important reference index for subsequent model optimization. By continuously reducing this total cost value, the analysis accuracy of the model for user requirements can be improved, thereby enhancing the performance and application effect of the model.

[0172] In one implementation, step S44 determines the requirement determination cost of the behavior data template library of the type to be classified based on the requirement determination cost and importance coefficient of each behavior data template of the type to be classified, including:

[0173] Step S441: Determine the average requirement determination cost of the first sub-library as the first requirement determination cost based on the importance coefficient and requirement determination cost of each behavior data template of the type to be classified in the first sub-library, where the first sub-library includes the behavior data templates of the type to be classified in the behavior data template library of the type to be classified whose requirement analysis values inferred by the first requirement analysis model are greater than the positive type threshold.

[0174] In one implementation, step S44 involves dividing the behavior data template library of the type to be classified into three sub-libraries, calculating the average requirement determination cost for each sub-library respectively, and finally obtaining the overall requirement determination cost through weighted summation.

[0175] Specifically, we first focus on the first sub-library, which contains the behavior data templates in the behavior data template library of the type to be classified, whose demand analysis values ​​inferred by the first demand analysis model are greater than the critical value of the positive type. For these behavior data templates, they usually represent the user's positive or strong demand expression. By calculating the product of the importance coefficient of each template in this sub-library and the demand determination cost, and finding the average, we get the first demand determination cost. This cost reflects the average error level of the model when predicting positive demand.

[0176] For example, suppose there are three templates A, B, and C in the first sub-library, and their importance coefficients are 0.5, 0.3, and 0.2, respectively, and their demand determination costs are 0.2, 0.3, and 0.1, respectively. Then, the calculation of the first demand determination cost is (0.5×0.2 + 0.3×0.3 + 0.2×0.1) / 3, which is the average of the weighted demand determination costs of all templates in this sub-library.

[0177] Based on this, step S441 provides a quantitative indicator, namely the first demand determination cost, which is used to evaluate the accuracy of the model in predicting positive demand. Subsequent steps S442 to S444 will use a similar method to calculate the demand determination cost for other sub-libraries, and finally derive the demand determination cost of the entire behavior data template library. This process helps to find out which demand types the model predicts with large errors and provides guidance for subsequent model optimization.

[0178] Step S442: Determine the average demand determination cost of the second sub-library based on the importance coefficient and the demand determination cost of each behavior data template of the to-be-classified type in the second sub-library as the second demand determination cost, wherein the second sub-library includes the behavior data template of the to-be-classified type in the behavior data template library whose demand analysis value inferred by the first demand analysis model is less than the negative type critical value.

[0179] Step S443: Determine the average demand determination cost of the third sub-library based on the demand determination cost of each behavior data template of the to-be-classified type in the third sub-library as the third demand determination cost, wherein the third sub-library includes the behavior data template of the to-be-classified type in the behavior data template library, whose demand analysis value inferred according to the first demand analysis model is not greater than the positive type critical value and not less than the negative type critical value.

[0180] Step S444: Determine the demand determination cost of the behavior data template library of the type to be classified based on the weighted sum of the first demand determination cost, the second demand determination cost and the third demand determination cost.

[0181] First, weights need to be determined for the cost determination of the first requirement, the cost determination of the second requirement, and the cost determination of the third requirement respectively. These weights can be set according to the actual situation. For example, if it is considered that positive requirements are more important than negative requirements, then the weight of the cost determination of the first requirement may be higher. Next, multiply the cost determination of each sub-library's requirement by its corresponding weight, and then add these three products together to obtain the weighted summation result. That is, the overall cost determination of the behavior data template library of the type to be classified.

[0182] It should be noted that the positive type threshold is the average result of the requirement analysis values indicating the positive type among the requirement analysis values of the behavior data templates of the type to be classified in the behavior data template library of the type to be classified; the negative type threshold is the average result of the requirement analysis values indicating the negative type among the requirement analysis values of the behavior data templates of the type to be classified in the behavior data template library of the type to be classified.

[0183] To determine which templates represent positive requirements and which represent negative requirements, two thresholds need to be set: the positive type threshold and the negative type threshold. These two thresholds are obtained by calculating the average value of the requirement analysis values of all templates in the library, but the calculation methods are different.

[0184] The calculation method of the positive type threshold is: First, screen out those requirement analysis values indicating the positive type from all templates (that is, those templates with higher values representing positive requirements); then, calculate the average value of these screened positive requirement analysis values, and the result obtained is the positive type threshold. This threshold represents the average requirement level of all positive requirement templates in the library. The calculation method of the negative type threshold is similar but in the opposite direction: First, screen out those requirement analysis values indicating the negative type (that is, those templates with lower values representing negative requirements); then, calculate the average value of these screened negative requirement analysis values, and the result obtained is the negative type threshold. This threshold represents the average requirement level of all negative requirement templates in the library.

[0185] Step S45: Optimize the model parameters of the first requirement analysis model based on the cost determination of the behavior data template library of the type to be classified.

[0186] Based on Step S45, the model can continuously self-adjust to more accurately predict and analyze user requirements. In this process, first, optimization algorithms (such as gradient descent algorithm, stochastic gradient descent algorithm, Adam optimization algorithm, etc.) can be used to adjust the model parameters according to the cost determination of the requirements. These parameters may include weights, biases, learning rates, etc., which together determine how the model processes and interprets the input behavior data.

[0187] The working principle of the optimization algorithm is to calculate the gradient of the cost with respect to the model parameters (i.e., the partial derivative of the cost function with respect to the parameters) based on the calculated requirements, and then update the parameters in the opposite direction of the gradient to gradually reduce the cost determined by the requirements. This process is repeated until the preset number of iterations is reached or the cost determined by the requirements is reduced to an acceptable range.

[0188] Assume that the model parameters are represented by the vector θ and the cost determined by the requirements is represented by the function J(θ). Then the optimization process can be mathematically expressed as:

[0189] θ = θ - α × ∇J(θ)

[0190] Where α is the learning rate, a positive decimal number used to control the step size of parameter update; ∇J(θ) is the gradient of the cost function J(θ) with respect to the parameter θ.

[0191] For example, if the cost determined by the requirements calculated in step S44 is 0.42 (this value may vary with different numbers of iterations and model states), and it is found that this cost is relatively high, indicating that the current parameter configuration of the model is not ideal and needs to be adjusted. In step S45, the above optimization algorithm will be used to update the model parameters based on this cost value, hoping to obtain a lower cost determined by the requirements in the next iteration, thereby improving the accuracy of the model's analysis of user requirements. Through such an iterative optimization process, the first requirement analysis model can gradually learn how to extract effective information from behavioral data and accurately predict user requirements, so as to provide more personalized and accurate services for users.

[0192] Regarding the fuzzy index mentioned in step S41, the method for obtaining it is introduced below. Specifically, in step S20, based on the errors between the requirement analysis values inferred by the first requirement analysis model and the second requirement analysis model for each behavior data template of the to-be-classified type in the behavior data template library, the fuzzy index of each behavior data template of the to-be-classified type is determined, including:

[0193] Step S21: Determine the modulus of the difference between the requirement analysis values inferred by the first requirement analysis model and the second requirement analysis model for each behavior data template of the to-be-classified type as the first error of each behavior data template of the to-be-classified type.

[0194] Specifically, for each behavior data template of the to-be-classified type, the first requirement analysis model and the second requirement analysis model will each output a requirement analysis value. These two values may be different because the two models may have learned different data features and patterns during the training process. Calculate the modulus of the difference between these two requirement analysis values, that is, take the absolute value of the difference between these two values, as the first error of the behavior data template of the to-be-classified type.

[0195] For example, assume that there is a behavior data template of the type to be classified regarding "booking a luxury suite". The demand analysis value of this template by the first demand analysis model is 0.8, while that of the second demand analysis model is 0.7. Then, according to step S21, calculate the modulus of the difference between these two values, that is, |0.8 - 0.7| = 0.1. 0.1 is the first error of the behavior data template of booking a luxury suite. The first error reflects the degree of consistency between the two models when conducting demand analysis on the behavior data template of the type to be classified. The larger the first error, the greater the difference in the demand analysis values of the two models, that is, the lower the consistency between the models; conversely, the smaller the first error, the closer the demand analysis values of the two models, and the higher the consistency between the models.

[0196] It should be noted that the modulus of the difference, that is, the absolute value, is calculated here because what is concerned is the magnitude of the difference between the two demand analysis values, rather than their signs or directions. Doing so helps to more accurately quantify the difference between the models and provides a reliable basis for calculating the fuzzy index in the subsequent steps.

[0197] Step S22: Determine the modulus of the sum value of the demand analysis values inferred by the first demand analysis model and the second demand analysis model respectively for each behavior data template of the type to be classified, as the second error of each behavior data template of the type to be classified.

[0198] Step S22 aims to further quantify the relationship between the demand analysis values obtained by the first demand analysis model and the second demand analysis model when inferring for the same behavior data template of the type to be classified. Different from step S21 which focuses on the difference between the two, step S22 focuses on the characteristics of their sum. Specifically, step S22 calculates the modulus of the sum value between the two demand analysis values. Here, the sum value refers to the sum of the two demand analysis values output by the first demand analysis model and the second demand analysis model for the same behavior data template of the type to be classified. For example, continuing with the behavior data template of booking a luxury suite of the type to be classified. Assume that the demand analysis value given by the first demand analysis model is 0.8, while the value given by the second demand analysis model is 0.7. According to step S22, we need to calculate the modulus of the sum value of these two values, that is, |0.8 + 0.7| = 1.5. 1.5 is the second error of the behavior data template of booking a luxury suite.

[0199] The second error reflects the comprehensive strength or overall tendency of the two models when conducting requirement analysis on the behavior data templates of the classification types. If the requirement analysis values of both models are relatively high, then their sum value will also be relatively large, and the second error will be correspondingly large, indicating that both models consider the behavior corresponding to this template to have a relatively high requirement; conversely, if the requirement analysis values of both models are relatively low, then the second error will be relatively small. It should be noted that although both step S21 and step S22 are calculated based on the requirement analysis values of the first requirement analysis model and the second requirement analysis model, they focus on different perspectives. Step S21 focuses on the difference between the two models, while step S22 focuses on the comprehensiveness or totality of the two models. These two steps together provide a basis for calculating the fuzzy index of the behavior data template of each classification type to be classified in the subsequent steps.

[0200] Step S23: Determine the fuzzy index of the behavior data template of each classification type to be classified, where the change trend of the fuzzy index of the behavior data template of each classification type to be classified is consistent with the change trend of the first error of the behavior data template of each classification type to be classified, and is opposite to the change trend of the second error of the behavior data template of each classification type to be classified.

[0201] In steps S21 and S22, the first error and the second error of the behavior data template of each classification type to be classified have been calculated. The first error represents the inconsistency of the requirement analysis values between the first requirement analysis model and the second requirement analysis model, while the second error represents the comprehensive strength of the requirement analysis values of both. In step S23, the fuzzy index is determined based on the first error and the second error. Specifically, the change trend of the fuzzy index should be consistent with the change trend of the first error and opposite to the change trend of the second error. This means that when the first error increases (i.e., the inconsistency between the models increases), the fuzzy index should also increase, indicating that the requirement analysis of the behavior data template of this classification type to be classified is more fuzzy or uncertain; conversely, when the second error increases (i.e., the comprehensive strength of the models increases), the fuzzy index should decrease because a higher comprehensive strength usually means that the models' requirement analysis of this template is more clear and consistent.

[0202]

[0203] Where E1 is the first error, E2 is the second error, and μ is a smoothing term used to prevent the denominator from becoming zero.

[0204] Please refer to Figure 3, which is a schematic structural diagram of a hotel service push device based on big data provided by an embodiment of the present application. The above-mentioned hotel service push device based on big data may be a computer program (including program code) running in a network device. For example, the hotel service push device based on big data is an application software; the device may be used to execute corresponding steps in the method provided by an embodiment of the present application. As Figure 3 shown, the hotel service push device based on big data may include: a data acquisition module 310, a model call module 320, a service push module 330, and a model debugging module 340.

[0205] Among them, the data acquisition module 310 is used to acquire big data on the behavior of the target user, and load the big data on the behavior of the target user into a hotel service demand analysis model that has been debugged in advance. Among them, the big data on the behavior of the target user includes hotel interaction behavior data of the target user; the model call module 320 is used to analyze the big data on the behavior of the target user according to the hotel service demand analysis model to obtain the hotel service demand of the target user; the service push module 330 is used to determine a target hotel service that matches the hotel service demand among preset hotel services according to the hotel service demand of the target user, and push the target hotel service to the client logged in by the target user; the model debugging module 340 is used to debug the hotel service demand analysis model. Among them, the hotel service demand analysis model is determined according to a behavior data template library, and the behavior data template library includes a behavior data template library of the positive type and a behavior data template library of the type to be classified. The hotel service demand analysis model is obtained through the following steps of debugging:

[0206] Configure the prior label of each behavior data template of the positive type into a positive type, configure the prior label of each behavior data template of the type to be classified into a negative type, and iteratively debug the hotel service demand analysis model with the behavior data template library as a training batch according to the prior label of each behavior data template, to obtain a first demand analysis model and the demand analysis value inferred by the first demand analysis model for each behavior data template of the type to be classified, and a second demand analysis model and the demand analysis value inferred by the second demand analysis model for each behavior data template of the type to be classified, where the first demand analysis model is obtained by iterative debugging for the first number of times, the second demand analysis model is obtained by iterative debugging for the second number of times, the first number is not equal to the second number, and in the iterative debugging, optimize the model parameter variables of the hotel service demand analysis model based on the demand determination cost of the behavior data template library; determine the fuzzy index of each behavior data template of the type to be classified based on the error between the demand analysis values inferred by the first demand analysis model and the second demand analysis model for each behavior data template of the type to be classified in the behavior data template library of the type to be classified; adjust the prior label of each behavior data template of the type to be classified based on the demand analysis value inferred by the first demand analysis model for each behavior data template of the type to be classified; iteratively debug the first demand analysis model with the behavior data template library of the type to be classified as a training batch according to the fuzzy index and the adjusted prior label of each behavior data template of the type to be classified, to obtain the debugged first demand analysis model; determine the model parameter variables of the hotel service demand analysis model according to the model parameter variables of the debugged first demand analysis model.

[0207] According to an embodiment of the present application, Figure 2 the steps involved in the big data-based hotel service push method shown can be Figure 3 executed by each module in the big data-based hotel service push device shown.

[0208] According to an embodiment of the present application, Figure 3Each module in the hotel service push device based on big data shown can be separately or all combined into one or several units to form, or a certain one (or some) of the units can be further split into at least two smaller sub-units in terms of function, and the same operations can be achieved without affecting the realization of the technical effects of the embodiments of this application. The above modules are divided based on logical functions. In practical applications, the function of one module can also be realized by at least two units, or the functions of at least two modules can be realized by one unit. In other embodiments of this application, the hotel service push device based on big data can also include other units. In practical applications, these functions can also be assisted by other units to be realized, and can be realized by the cooperation of at least two units.

[0209] According to an embodiment of this application, it can be achieved by running a computer program (including program code) capable of executing the respective steps involved in the corresponding method shown in Figure 2 on a general computer device such as a computer including processing components and storage components such as a central processing unit (CPU), a random access storage medium (RAM), and a read-only storage medium (ROM), to construct the hotel service push device based on big data shown in Figure 3 and to implement the hotel service push method based on big data of the embodiments of this application. The above computer program can be recorded on a computer-readable recording medium, for example, and loaded into the above computing device through the computer-readable recording medium and run therein. Please refer to Figure 4 , which is a schematic structural diagram of a computer device provided by an embodiment of this application. As shown in Figure 4 , the above computer device 1000 can include: a processor 1001, a network interface 1004, and a memory 1005. In addition, the above computer device 1000 can also include: a user interface 1003, and at least one communication bus 1002. Among them, the communication bus 1002 is used to realize the connection communication between these components. Among them, the user interface 1003 can include a display screen (Display) and a keyboard (Keyboard). Optionally, the user interface 1003 can also include a standard wired interface and a wireless interface. The network interface 1004 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a non-volatile memory, for example, at least one disk memory. Optionally, the memory 1005 can also be at least one storage device far from the aforementioned processor 1001. As shown in Figure 4 , the memory 1005, as a computer-readable storage medium, can include an operating system, a network communication module, a user interface module, and a device control application program.

[0210] InFigure 4 In the computer device 1000 shown, the network interface 1004 can provide network communication functions; the user interface 1003 is mainly used to provide an input interface; and the processor 1001 can be used to call the device control application program stored in the memory 1005 to implement the methods provided in the above embodiments. It should be understood that the computer device 1000 described in the embodiments of the present application can execute the Figure 2 description of the method for pushing hotel services based on big data in the corresponding previous embodiments, and can also execute the Figure 3 description of the device for pushing hotel services based on big data in the corresponding previous embodiments, which will not be elaborated here. In addition, the description of the beneficial effects of using the same method will not be elaborated either.

[0211] In addition, it should be noted here that: The embodiments of the present application also provide a computer-readable storage medium, and the above computer-readable storage medium stores the computer program executed by the device for pushing hotel services based on big data mentioned above, and the above computer program includes program instructions. When the above processor executes the above program instructions, it can execute the Figure 2 description of the method for pushing hotel services based on big data in the corresponding previous embodiments. Therefore, it will not be elaborated here. In addition, the description of the beneficial effects of using the same method will not be elaborated either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application.

[0212] As an example, the above program instructions can be deployed to be executed on a computer device, or deployed to be executed on at least two computer devices at one location. Or, they can be executed on at least two computer devices distributed at at least two locations and interconnected through a communication network. The at least two computer devices distributed at at least two locations and interconnected through a communication network can form a blockchain network.

[0213] The above computer-readable storage medium may be the middle storage unit of the hotel service push device based on big data provided in any of the foregoing embodiments or the above computer device, such as the hard disk or middle 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 media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the computer-readable storage medium may also include both the middle storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0214] In the description of the specification, claims, and drawings of the embodiments of the present application, the terms "first", "second", etc. are used to distinguish different contents, rather than to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units is not limited to the listed steps or modules, but may optionally further include steps or modules not listed, or may optionally further include other step units inherent to these processes, methods, devices, products, or equipment.

[0215] The embodiments of the present application also provide a computer program product, including a computer program / instructions, where when the computer program / instructions are executed by a processor, they implement the description of the above-mentioned hotel service push method based on big data in the corresponding embodiments Figure 2 described above, and therefore, details will not be repeated here. In addition, the description of the beneficial effects of adopting the same method will not be repeated either. For the technical details not disclosed in the embodiments of the computer program product involved in the present application, please refer to the description of the method embodiments of the present application.

Claims

1. A hotel service push method based on big data, characterized in that: Applied to a computer device, the method comprises: Obtaining target user behavior big data, and loading the target user behavior big data into a pre-debugged hotel service demand analysis model, wherein the target user behavior big data includes the hotel interaction behavior data of the target user; Analyze the target user behavior big data according to the hotel service demand analysis model to obtain the hotel service demand of the target user; According to the hotel service demand of the target user, determine the target hotel service matching the hotel service demand from the preset hotel services, and push the target hotel service to the client where the target user logs in; The hotel service demand analysis model is determined based on a behavior data template library, wherein the behavior data template library includes a positive type behavior data template library and a to-be-classified type behavior data template library, and the hotel service demand analysis model is debugged through the following steps: The priori mark of each positive type of behavior data template is configured as a positive type, and the priori mark of each behavior data template of the type to be classified is configured as a negative type, and according to the priori mark of each behavior data template, the hotel service demand analysis model is iteratively debugged with the behavior data template library as a training batch to obtain the first demand analysis model and the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified, and the second demand analysis model and the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified, wherein the first demand analysis model is obtained by iterative debugging using a first number, and the second demand analysis model is obtained by iterative debugging using a second number, the first number is not equal to the second number, and in the iterative debugging, the model parameter of the hotel service demand analysis model is optimized based on the demand determination cost of the behavior data template library; Determine the fuzzy index of each behavior data template of the to-be-classified type based on the errors between the demand analysis values ​​inferred by the first demand analysis model and the second demand analysis model for each behavior data template of the to-be-classified type in the behavior data template library of the to-be-classified type; Adjusting the priori label of each type of behavior data template to be classified based on the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified; Determine the importance coefficient of each type of behavior data template to be classified based on the fuzzy index of each type of behavior data template to be classified, and the importance coefficient has an opposite trend to the change of the fuzzy index; repeatedly perform the following operations until the first demand analysis model reaches the debugging cutoff condition to obtain the debugged first demand analysis model: determine the demand determination value of each type of behavior data template to be classified based on the first demand analysis model; determine the demand determination cost of each type of behavior data template to be classified based on the prior label and the demand determination value of each type of behavior data template to be classified; determine the demand determination cost of the behavior data template library of the type to be classified based on the demand determination cost and the importance coefficient of each type of behavior data template to be classified; optimize the model parameters of the first demand analysis model based on the demand determination cost of the behavior data template library of the type to be classified; The model parameters of the hotel service demand analysis model are determined according to the model parameters of the debugged first demand analysis model.

2. The method according to claim 1, characterized in that The step of adjusting the prior label of each behavior data template of the to-be-classified type based on the demand analysis value inferred by the first demand analysis model for each behavior data template of the to-be-classified type includes: For each behavior data template of the to-be-classified type in the behavior data template library of the to-be-classified type, the following operations are performed: When the demand analysis value inferred by the first demand analysis model for the behavior data template of the type to be classified is greater than the positive type critical value, adjusting the priori mark of the behavior data template of the type to be classified to the positive type; When the demand analysis value inferred by the first demand analysis model for the behavior data template of the type to be classified is not greater than the positive type critical value, the priori label of the behavior data template of the type to be classified is adjusted to a negative type.

3. The method according to claim 1, characterized in that The fuzzy index of each behavior data template of the to-be-classified type is not less than zero, and the importance coefficient of each behavior data template of the to-be-classified type is determined based on the fuzzy index of each behavior data template of the to-be-classified type, including: The ratio of the opposite number of the fuzzy index of each type of behavior data template to be classified to the first preset model parameter is used as the index of the preset value to determine the importance coefficient of each type of behavior data template to be classified.

4. The method according to claim 1, characterized in that: The step of determining the demand determination cost of the behavior data template library of the type to be classified according to the demand determination cost and importance coefficient of each behavior data template of the type to be classified includes: Determine the average demand determination cost of the first sub-library based on the importance coefficient and demand determination cost of each behavior data template of the to-be-classified type in the first sub-library as the first demand determination cost, wherein the first sub-library includes the behavior data templates of the to-be-classified type whose demand analysis values ​​inferred by the first demand analysis model are greater than the positive type critical value in the behavior data template library of the to-be-classified type; Determine, based on the importance coefficient of each behavior data template of the to-be-classified type in the second sub-library and the demand determination cost, an average of the demand determination costs of the second sub-library as the second demand determination cost, wherein the second sub-library includes the behavior data templates of the to-be-classified type whose demand analysis values ​​inferred by the first demand analysis model are less than the negative type critical value in the behavior data template library of the to-be-classified type; Determine an average of the demand determination costs of the third sub-library based on the demand determination costs of each behavior data template of the to-be-classified type in the third sub-library as a third demand determination cost, wherein the third sub-library includes the behavior data templates of the to-be-classified type in the behavior data template library of the to-be-classified type, whose demand analysis values ​​inferred according to the first demand analysis model are not greater than the positive type critical value and not less than the negative type critical value; Determining the demand determination cost of the behavior data template library of the type to be classified based on a weighted sum of the first demand determination cost, the second demand determination cost and the third demand determination cost; Among them, the positive type critical value is the average result of each demand analysis value indicating the positive type in the demand analysis values ​​of the behavior data template to be classified in the behavior data template library of the type to be classified; the negative type critical value is the average result of each demand analysis value indicating the negative type in the demand analysis values ​​of the behavior data template to be classified in the behavior data template library of the type to be classified.

5. The method according to any one of claims 1 to 4, characterized in that The method of iteratively debugging the hotel service demand analysis model based on the prior label of each behavior data template and taking the behavior data template library as a training batch, obtaining the first demand analysis model and the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified, and the second demand analysis model and the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified, includes: In the first number of iterations of the hotel service demand analysis model, the demand determination value for each behavior data template of the to-be-classified type in the last iteration is determined as the demand analysis value inferred by the first demand analysis model for each behavior data template of the to-be-classified type; In the second number of iterations of the hotel service demand analysis model, the demand determination value for each behavior data template of the type to be classified obtained in the last iteration is determined as the demand analysis value inferred by the second demand analysis model for each behavior data template of the type to be classified; Alternatively, the hotel service demand analysis model is iteratively debugged based on the prior label of each behavior data template and the behavior data template library is used as a training batch to obtain the first demand analysis model and the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified, and the second demand analysis model and the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified, including: According to the pre-deployed eccentricity adjustment coefficient, determine the weighted sum of the various demand determination values ​​output by the hotel service demand analysis model for each type of behavior data template to be classified in the first number of iterative debugging, as the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified; According to the pre-deployed eccentricity adjustment coefficient, determine the weighted sum of the various demand determination values ​​output by the hotel service demand analysis model for each type of behavior data template to be classified in the second number of iterative debugging, as the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified; Alternatively, the hotel service demand analysis model is iteratively debugged based on the prior label of each behavior data template and the behavior data template library is used as a training batch to obtain the first demand analysis model and the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified, and the second demand analysis model and the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified, including: For each behavior data template to be classified, perform the following operations: The result of multiplying the demand determination value of the behavior data template of the type to be classified in the first iteration debugging of the hotel service demand analysis model by a second preset model parameter variable is used as the demand analysis value of the behavior data template of the type to be classified in the first iteration debugging of the hotel service demand analysis model, and the second preset model parameter variable is between 0 and 1; The demand analysis value of the behavior data template of the to-be-classified type in the other iterative debugging reasoning of the hotel service demand analysis model is determined based on the following formula: h n (i)=a·h m (i)+b·f n (i), Where i represents the behavior data template of the type to be classified, n represents the round number of iterative debugging, and h n (i) is the demand analysis value of the behavior data template of the type to be classified in the nth iteration debugging reasoning of the hotel service demand analysis model, h m (i) is the demand analysis value of the behavior data template of the type to be classified in the mth iteration debugging reasoning of the hotel service demand analysis model, a and b are the second preset model parameters, f n (i) determining a demand value for the behavior data template of the type to be classified in the nth iteration debugging of the hotel service demand analysis model; wherein b=1-a, m=n-1; Using the demand analysis value of the behavior data template of the type to be classified in the first number of iterations of the hotel service demand analysis model as the demand analysis value inferred by the first demand analysis model for the behavior data template of the type to be classified; The demand analysis value of the behavior data template of the type to be classified in the second number of iterative debugging of the hotel service demand analysis model is used as the demand analysis value inferred by the second demand analysis model for the behavior data template of the type to be classified.

6. The method according to any one of claims 1 to 4, characterized in that The step of determining the fuzzy index of each behavior data template of the to-be-classified type based on the error between the demand analysis values ​​inferred by the first demand analysis model and the second demand analysis model for each behavior data template of the to-be-classified type in the behavior data template library of the to-be-classified type comprises: Determine the modulus of the difference between the demand analysis values ​​inferred by the first demand analysis model and the second demand analysis model for each type of behavior data template to be classified, as the first error of each type of behavior data template to be classified; Determine the modulus of the sum of the demand analysis values ​​inferred by the first demand analysis model and the second demand analysis model for each type of behavior data template to be classified, as the second error of each type of behavior data template to be classified; Determine a fuzzy index of each type of behavior data template to be classified, wherein the fuzzy index of each type of behavior data template to be classified is consistent with a changing trend of a first error of the behavior data template to be classified, and is opposite to a changing trend of a second error of the behavior data template to be classified.

7. The method according to any one of claims 1 to 4, characterized in that Determining the model parameter variables of the hotel service demand analysis model according to the model parameter variables of the debugged first demand analysis model includes: Determine the model parameter variables of the debugged first demand analysis model as the model parameter variables of the hotel service demand analysis model; Alternatively, determining the model parameters of the hotel service demand analysis model based on the model parameters of the debugged first demand analysis model includes: Adjusting the priori label of each type of behavior data template to be classified based on the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified; According to the fuzzy index of each behavior data template of the to-be-classified type and the adjusted priori mark, the second demand analysis model is iteratively debugged with the behavior data template library of the to-be-classified type as a training batch to obtain the debugged second demand analysis model; For each model parameter of the debugged first demand analysis model, an average result between the model parameter and the corresponding model parameter of the second demand analysis model is determined as the corresponding model parameter of the hotel service demand analysis model.

8. A hotel service push device based on big data, characterized in that: include: A data acquisition module is used to acquire target user behavior big data, and load the target user behavior big data into a pre-debugged hotel service demand analysis model, wherein the target user behavior big data includes the hotel interaction behavior data of the target user; A model calling module is used to analyze the target user behavior big data according to the hotel service demand analysis model to obtain the hotel service demand of the target user; A service push module is used to determine a target hotel service matching the hotel service demand from preset hotel services according to the hotel service demand of the target user, and push the target hotel service to the client where the target user logs in; A model debugging module is used to debug the hotel service demand analysis model, wherein the hotel service demand analysis model is determined according to a behavior data template library, the behavior data template library includes a positive type behavior data template library and a to-be-classified type behavior data template library, and the hotel service demand analysis model is debugged by the following steps: The priori mark of each positive type of behavior data template is configured as a positive type, and the priori mark of each behavior data template of the type to be classified is configured as a negative type, and according to the priori mark of each behavior data template, the hotel service demand analysis model is iteratively debugged with the behavior data template library as a training batch to obtain the first demand analysis model and the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified, and the second demand analysis model and the demand analysis value inferred by the second demand analysis model for each type of behavior data template to be classified, wherein the first demand analysis model is obtained by iterative debugging using a first number, and the second demand analysis model is obtained by iterative debugging using a second number, the first number is not equal to the second number, and in the iterative debugging, the model parameter of the hotel service demand analysis model is optimized based on the demand determination cost of the behavior data template library; Determine the fuzzy index of each behavior data template of the to-be-classified type based on the errors between the demand analysis values ​​inferred by the first demand analysis model and the second demand analysis model for each behavior data template of the to-be-classified type in the behavior data template library of the to-be-classified type; Adjusting the priori label of each type of behavior data template to be classified based on the demand analysis value inferred by the first demand analysis model for each type of behavior data template to be classified; Determine the importance coefficient of each type of behavior data template to be classified based on the fuzzy index of each type of behavior data template to be classified, and the importance coefficient has an opposite trend to the change of the fuzzy index; repeatedly perform the following operations until the first demand analysis model reaches the debugging cutoff condition to obtain the debugged first demand analysis model: determine the demand determination value of each type of behavior data template to be classified based on the first demand analysis model; determine the demand determination cost of each type of behavior data template to be classified based on the prior label and the demand determination value of each type of behavior data template to be classified; determine the demand determination cost of the behavior data template library of the type to be classified based on the demand determination cost and the importance coefficient of each type of behavior data template to be classified; optimize the model parameters of the first demand analysis model based on the demand determination cost of the behavior data template library of the type to be classified; The model parameters of the hotel service demand analysis model are determined according to the model parameters of the debugged first demand analysis model.

9. A computer device, characterized in that: include: processor; and a memory, wherein the memory stores a computer-readable storage medium, and when the computer-readable storage medium is executed by the processor, the processor executes the method as claimed in any one of claims 1 to 7.

Images