Hotel order fuzzy matching method based on large model

Abstract

The invention discloses a hotel order fuzzy matching method based on a large model. The method comprises the following steps of 1) collecting hotel order screenshot information of each platform by using a hotel order screenshot mode; 2) constructing an order keyword library based on the hotel order screenshot information; 3) training a graph neural network model by using the order keyword library; 4) acquiring screenshot information uploaded by the user; 5) establishing a fuzzy membership function by using the graph neural network model, and optimizing the fuzzy membership function by using the loss function; and performing semantic supplement on the screenshot information by using the optimized fuzzy membership function, and writing the supplemented semantic features into the semantic content extracted in the step 5) to generate the semantic content without missing. According to the method, semantic rule understanding is realized by calling a large model interface, and meanwhile, fuzzy matching is performed on orders of hotels by applying a fuzzy membership function, so that quick matching of hotel screenshots and hotels is realized.

Description

Technical Field

[0001] This invention relates to the field of hotel order text recognition, specifically a fuzzy matching method for hotel orders based on a large model. Background Technology

[0002] With the continuous evolution of artificial intelligence technology and the booming development of the online hotel booking industry, various hotel booking platforms are proliferating. However, complex pricing strategies and "price discrimination based on big data" are common occurrences among these platforms, often requiring users to search and compare prices on multiple platforms simultaneously to find affordable and cost-effective hotels. In this process, users commonly take screenshots of the order information pages displayed on their mobile phones or computers to capture and temporarily save real-time prices and room details from different platforms.

[0003] However, in practice, due to the limited screen size of mobile phones or the complex layout of web pages / app interfaces, the order images manually captured by users often have incomplete content and missing key information.

[0004] Currently, the industry typically uses OCR (Optical Character Recognition) technology to automatically recognize text information in screenshots. However, due to the inherent information gaps in screenshots, OCR technology can only mechanically recognize the actual text present in the image, failing to understand the semantic relationships between the truncated information. For example, it cannot determine whether prices and room types in two different screenshots belong to the same order. This lack of text due to incomplete screenshots severely hinders the accurate extraction and structured processing of hotel order information, making it difficult for the system to achieve effective multi-platform data integration and price comparison. Summary of the Invention

[0005] The purpose of this invention is to provide a fuzzy matching method for hotel orders based on a large model, comprising the following steps:

[0006] Step 1) Collect hotel order screenshots from various platforms;

[0007] Step 2) Based on hotel order screenshots, build an order keyword database;

[0008] Step 3) Train a graph neural network model using the order keyword database;

[0009] Step 4) Obtain screenshot information uploaded by the user;

[0010] Step 5) Use large models and OCR technology to perform semantic understanding on the screenshot information, extract semantic content, and determine whether there are missing semantic features in the semantic content;

[0011] If there are any missing parts, proceed to step 6).

[0012] Step 6) Establish a fuzzy membership function using a graph neural network model, and optimize the fuzzy membership function using a loss function;

[0013] The optimized fuzzy membership function is used to semantically supplement the screenshot information, and the supplemented semantic features are written into the semantic content extracted in step 5) to generate semantic content without missing features.

[0014] Furthermore, in step 1), the platform is a platform that provides hotel booking services.

[0015] In step 1), the hotel order screenshot includes the order details page and the payment details page.

[0016] Furthermore, in step 2), when constructing the order keyword library, each field is treated as a node in a graph neural network, and the correlation and semantic information between words are set as edges, with the weight of the edges being semantic similarity. ;

[0017] Among them, semantic similarity As shown below:

[0018] (1)

[0019] In the formula, , For different semantic content.

[0020] Furthermore, in step 3), the order keyword database includes hotel name, room type name, check-in time, whether breakfast is included, contact person for booking, hotel price, points program, mileage redemption, promotional activities, coupons / vouchers, and room policy.

[0021] Furthermore, when training a graph neural network, each node of the graph neural network... In the Layer representation The update is as follows:

[0022] (2)

[0023] In the formula, For nodes In the Layer representation, For nodes Its neighboring nodes Edge weights between them; For nodes The set of neighboring nodes, This is the activation function. As weight.

[0024] Furthermore, in step 5), the large model is an AI large model.

[0025] Furthermore, in step 5), the semantic content of each field is extracted as follows:

[0026] (3)

[0027] In the formula, For semantic content; For fields.

[0028] Furthermore, in step 6), for each field The fuzzy membership function is shown below:

[0029] (4)

[0030] in: For fields Fuzzy membership degree, and These are the weights and biases used in fuzzy membership calculation. It's the Sigmoid function: .

[0031] Furthermore, in step 6), the loss function is as follows:

[0032] (5)

[0033] In the formula, This represents the true membership degree. This is the loss function.

[0034] Furthermore, complete semantic content is used to search for and match hotel room types in the order keyword library.

[0035] The technical effectiveness of this invention is undeniable. By introducing large-scale model semantic understanding and fuzzy membership functions, this invention solves the problem of recognition and matching caused by incomplete order screenshot information.

[0036] Specifically, the beneficial effects of the present invention are as follows:

[0037] 1) This invention uses a large model interface to perform semantic rule understanding and contextual analysis on fragmented text. It can intelligently infer and complete missing key order fields, thereby reconstructing complete and structured order information from a set of incomplete screenshots, fundamentally solving the problem of missing text caused by incomplete screenshots.

[0038] 2) To address potential errors or uncertainties in the completed order information, this invention employs a fuzzy membership function to perform comprehensive similarity calculations on multiple attributes such as hotel name, room type characteristics, and price range. This method exhibits strong tolerance for minor errors (such as typos and aliases) and missing information in text recognition, enabling rapid and accurate matching of target hotels and their room types within a large-scale hotel database, significantly improving the recall and accuracy of order matching.

[0039] 3) This invention can automatically complete the query, alignment and comparison of cross-platform data, and present the comparison results intuitively, reducing the operation steps and time costs required for users to obtain comprehensive information. Attached Figure Description

[0040] Figure 1 This is a flowchart of the algorithm. Detailed Implementation

[0041] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0042] Example 1:

[0043] See Figure 1 A fuzzy matching method for hotel orders based on a large model includes the following steps:

[0044] Step 1) Collect hotel order screenshots from various platforms;

[0045] Step 2) Based on hotel order screenshots, build an order keyword database;

[0046] Step 3) Train a graph neural network model using the order keyword database;

[0047] Step 4) Obtain screenshot information uploaded by the user;

[0048] Step 5) Use large models and OCR technology to perform semantic understanding on the screenshot information, extract semantic content, and determine whether there are missing semantic features in the semantic content;

[0049] If there are any missing parts, proceed to step 6).

[0050] Step 6) Establish a fuzzy membership function using a graph neural network model, and optimize the fuzzy membership function using a loss function;

[0051] The optimized fuzzy membership function is used to semantically supplement the screenshot information, and the supplemented semantic features are written into the semantic content extracted in step 5) to generate semantic content without missing features.

[0052] Example 2:

[0053] A fuzzy matching method for hotel orders based on a large model, with the same technical content as in Embodiment 1, further wherein, in step 1), the platform is a platform that provides hotel booking services.

[0054] In step 1), the hotel order screenshot includes the order details page and the payment details page.

[0055] Example 3:

[0056] A fuzzy matching method for hotel orders based on a large model, with technical content identical to any one of embodiments 1-2, further comprising the following step 2): when constructing the order keyword database, each field is treated as a node in a graph neural network, and the correlation and semantic information between words are set as edges, with the weight of the edges being semantic similarity. ;

[0057] Among them, semantic similarity As shown below:

[0058] (1)

[0059] In the formula, , For different semantic content.

[0060] Example 4:

[0061] A fuzzy matching method for hotel orders based on a large model, with the same technical content as any one of embodiments 1-3, further wherein, in step 3), the order keyword library includes hotel name, room type name, check-in time, whether breakfast is included, booking contact person, hotel price, points program, mileage redemption, promotional activities, coupons, and room policies.

[0062] Example 5:

[0063] A fuzzy matching method for hotel orders based on a large model, with technical content the same as any one of embodiments 1-4, further comprising the following steps: when training the graph neural network, each node of the graph neural network... In the Layer representation The update is as follows:

[0064] (2)

[0065] In the formula, For nodes In the Layer representation, For nodes Its neighboring nodes Edge weights between them; For nodes The set of neighboring nodes, This is the activation function. As weight.

[0066] Example 6:

[0067] A fuzzy matching method for hotel orders based on a large model, the technical content of which is the same as any one of embodiments 1-5, further wherein, in step 5), the large model is an AI large model.

[0068] Example 7:

[0069] A fuzzy matching method for hotel orders based on a large model, with the same technical content as any one of embodiments 1-6, further wherein, in step 5), the semantic content of each field is extracted as follows:

[0070] (3)

[0071] In the formula, For semantic content; For fields. BERT is a pre-trained language model.

[0072] Example 8:

[0073] A fuzzy matching method for hotel orders based on a large model, with technical content the same as any one of embodiments 1-7, further wherein, in step 6), for each field... The fuzzy membership function is shown below:

[0074] (4)

[0075] in: For fields Fuzzy membership degree, and These are the weights and biases used in fuzzy membership calculation. It's the Sigmoid function: .

[0076] Example 9:

[0077] A fuzzy matching method for hotel orders based on a large model, with the same technical content as any one of embodiments 1-8, further wherein, in step 6), the loss function is as follows:

[0078] (5)

[0079] In the formula, This represents the true membership degree. This is the loss function.

[0080] Example 10:

[0081] A fuzzy matching method for hotel orders based on a large model is provided, with the technical content being the same as any one of embodiments 1-9. Furthermore, semantic content without missing information is used to search for and match hotel room types in the order keyword library.

[0082] Example 11:

[0083] A fuzzy matching method for hotel orders based on a large model, comprising the following steps:

[0084] 1) Collect order screenshots from major platforms, including Ctrip, Meituan, Qunar, Tongcheng, Fliggy, Zhixing, eLong, Douyin, JD.com, Tuniu, and 10 other platforms.

[0085] 2) Construct an order keyword library and treat each field as a node in a graph neural network, setting the relationships and semantic information between words as edges.

[0086] 3) Train the graph neural network model.

[0087] 4) For the uploaded screenshot information, use large model and OCR technology to extract and understand the semantics of the screenshot information and identify the semantic content.

[0088] 5) Establish a fuzzy membership function for the missing semantic parts.

[0089] 6) Set cross-entropy as the loss function and train the model.

[0090] 7) Using the calculated fuzzy membership degree and graph neural network node features, the missing part is inferred from the context.

[0091] 8) Based on the reasoning results, quickly search and match hotel room types in the database (i.e., the keyword database).

[0092] Order data from 10 platforms, including Ctrip, Meituan, Qunar, Tongcheng, Fliggy, Zhixing, eLong, Douyin, JD.com, and Tuniu, was collected using screenshots. This data mainly includes, but is not limited to, order details pages and payment details pages. The specific pages were determined based on the keywords described in claim 3.

[0093] The keyword database refers to the 12 key information elements that the order screenshot page must contain, or should contain as many as possible, as possible: hotel name, room type name, check-in time, whether breakfast is included, contact person for booking, hotel price, points program, mileage redemption, promotional activities, coupons / vouchers, and room policies.

[0094] The screenshots are semantically understood using a large model, which includes, but is not limited to, the Doubao multimodal general model and the Tongyi multimodal model. The OCR technology for extracting text from the screenshots includes, but is not limited to, EasyOCR technology.

[0095] Using a large model and OCR technology, semantic vectors for each field in hotel orders are extracted. .

[0096] For each field Inputting this into a large model (Doubao multimodal general model, Tongyi multimodal model) generates semantic vectors. .

[0097]

[0098] Based on the collected keyword database, each field of the hotel order is represented as a node in a graph, and the edges in the graph connect the relationships between the nodes (the weights of the edges are mainly determined by the order of the fields and the semantic relevance).

[0099] The node set is edge set is The weight of the edge is the semantic similarity. The calculation formula is as follows:

[0100]

[0101] This invention constructs a graph structure, forming a graph-structured database of hotel order corpus.

[0102] For training graph neural networks, this invention mainly uses graph neural networks for message passing, updating the representation of each node through the features of neighboring nodes.

[0103] For each node , which means In the The update formula for the layer is:

[0104]

[0105] in:

[0106] For nodes In the Layer representation,

[0107] For the first The weight matrix of the layer,

[0108] For nodes Its neighboring nodes Edge weights between them

[0109] For nodes The set of neighboring nodes,

[0110] For activation functions (such as ReLU)

[0111] The membership function is constructed by calculating the fuzzy membership degree of each field based on the updated node representation of the graph neural network, which represents the degree of missingness of the field.

[0112] For each field Calculate its fuzzy membership degree:

[0113]

[0114] in:

[0115] For fields Fuzzy membership degree,

[0116] and These are the weights and biases used in fuzzy membership calculation.

[0117] It's the Sigmoid function: .

[0118] We set up a cross-entropy loss function to optimize the calculation of fuzzy membership degrees, hoping that the model can accurately predict the degree of field missingness.

[0119]

[0120] This invention combines OCR technology with general large model technology. Users only need to upload screenshots to immediately compare prices across platforms. At the same time, based on existing database information and hotel room types obtained from the user's uploaded screenshots, we will recommend the lowest price and discount, reducing the time users spend comparing prices across multiple platforms.

[0121] This invention helps increase hotel sales, eliminating the need for hotels to compare prices across multiple platforms, and allowing users to obtain the maximum discount, achieving a win-win situation for both hotels and customers.

[0122] In summary, this invention, based on AI algorithms, realizes an autonomous negotiation model that benefits customers while ensuring hotel profits, achieving a win-win situation for both customers and hotels.

[0123] Example 12:

[0124] A fuzzy matching method for hotel orders based on a large model is disclosed. The technical content is the same as any one of embodiments 1-11. Further, this invention mainly uses hotel order screenshots to achieve rapid matching and recommendation of hotel room types. Order screenshots are collected from various platforms including Ctrip, Meituan, Qunar, Tongcheng, Fliggy, Zhixing, eLong, Douyin, JD.com, and Tuniu (10 platforms in total). This invention primarily uses order screenshots for tagging and training. Order screenshots include, but are not limited to, order details pages and payment details pages. Only screenshots containing 12 elements of information are needed, such as hotel name, room type name, check-in time, whether breakfast is included, booking contact person, hotel price, points program, mileage redemption, promotional activities, coupons, and room policies. All such screenshots belong to the database for training the tagging model of this invention.

[0125] Example 13:

[0126] A fuzzy matching method for hotel orders based on a large model is disclosed. The technical content is the same as any one of embodiments 1-12. Further, in the embodiments, the present invention first uses OCR technology to obtain hotel order information from the uploaded hotel screenshot data, and then uses a large model to perform semantic understanding on the recognized text content. This large model mainly includes the Doubao general large model and the general semantic large model, which mainly realize the extraction of semantic text and use it as training input data for GNN and fuzzy membership degree.

[0127] Example 14:

[0128] A fuzzy matching method for hotel orders based on a large model is disclosed. The technical content is the same as any one of embodiments 1-13. Further, during the GNN invocation process, this invention represents each field of the hotel order as a node in a graph, with edges connecting the relationships between nodes. The definition of a node can be keywords of 12 elements (hotel name, room type name, check-in time, breakfast included / not included, booking contact person, hotel price, points program, mileage redemption, promotional activities, coupons / vouchers, room policies, etc.), not just limited to a single field as a node; treating keywords as nodes is still within the scope of protection. Furthermore, the relationships between edges are not limited to a single keyword field; they can be described as the connection between a keyword and that field, as well as the connections between fields.

Claims

1. A large model-based hotel order fuzzy matching method, characterized in that, Includes the following steps: Step 1) Collect hotel order screenshots from various platforms; Step 2) Based on hotel order screenshots, build an order keyword database; Step 3) Train a graph neural network model using the order keyword database; Step 4) Obtain screenshot information uploaded by the user; Step 5) Use large models and OCR technology to perform semantic understanding on the screenshot information, extract semantic content, and determine whether there are missing semantic features in the semantic content; If there are any missing parts, proceed to step 6). Step 6) Establish a fuzzy membership function using a graph neural network model, and optimize the fuzzy membership function using a loss function; The optimized fuzzy membership function is used to semantically supplement the screenshot information, and the supplemented semantic features are written into the semantic content extracted in step 5) to generate semantic content without missing features.

2. The method according to claim 1, wherein, In step 1), the platform is a platform that provides hotel booking services; In step 1), the hotel order screenshot includes the order details page and the payment details page.

3. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, In step 2), when constructing the order keyword library, each field is regarded as a node of a graph neural network, and the relevance and semantic information between words are set as edges, and the weight of the edge is the semantic similarity ; wherein the semantic similarity as follows: ；（1） In the formula, , are different semantic contents.

4. The method according to claim 1, wherein, In step 3), the order keyword database includes hotel name, room type name, check-in time, whether breakfast is included, contact person for booking, hotel price, points program, mileage redemption, promotional activities, coupons and vouchers, and room policy.

5. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, When training the graph neural network, the graph neural network updates each node In the first representation of the layer is updated as follows: ；（2） wherein is a node In a first representation of a layer, is a node and its neighbor nodes between them; is a set of neighbor nodes of a node is an activation function; is a weight.

6. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, In step 5), the large model is an AI large model.

7. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, In step 5), the semantic content of each field is extracted as follows: ；（3） In the formula, For semantic content; For fields.

8. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, In step 6), for each field The fuzzy membership function is shown below: ；（4） in: For fields Fuzzy membership degree, and These are the weights and biases used in fuzzy membership calculation. It's the Sigmoid function: .

9. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that, In step 6), the loss function is as follows: ； （5） In the formula, The actual membership degree; This is the loss function.

10. The fuzzy matching method for hotel orders based on a large model according to claim 1, characterized in that: Complete semantic content is used to search and match hotel room types in the order keyword library.

Images