An API gateway recommendation method and system based on multimodal feature fusion

Abstract

The present invention relates to the technical field of API gateway recommendation, and specifically discloses a multimodal feature fusion API gateway recommendation method and system, the method comprising: step 1, receiving a client service request authentication at an API gateway entrance, accessing multimodal data through a multi-protocol proxy module to a route, and requesting a filtering plug-in to perform preliminary verification; step 2, receiving an image processing service request and a text processing service request at a receiving layer, extracting features from the multimodal data after preliminary verification, and storing the features; step 3, receiving call information at a feature layer, fusing features of the same processing node, confirming the fusion strategy and sending it to a decision layer; step 4, generating an API gateway recommendation based on the fusion strategy, and generating a recommendation engine feedback output to a client; the present invention realizes the coordinated fusion of multimodal data features and API gateway architecture information, and promotes the optimization of the API gateway recommendation process.

Description

Technical Field

[0001] The present invention relates to the field of API gateway recommendation technology, and in particular to a multimodal feature fusion API gateway recommendation method and system. Background Art

[0002] An API gateway is an entry point for accessing and querying a system, responsible for request routing, security control, load balancing, and other functions. In particular, when recommending an API gateway, it is usually necessary to perform entry access verification on various multi-source data types to ensure the security of data input. Therefore, as a traffic entry point, the API gateway needs to handle high-concurrency requests, resulting in a surge in traffic load, which can easily cause single-point gateway failures and lead to crashes.

[0003] Through the multimodal feature fusion process to integrate different data types, and combined with the architectural characteristics of the API gateway to design the corresponding request service recommendation method, to ensure that the needs of different services are achieved through the gateway according to the user request information, and to return and respond to the recommendation results. Summary of the Invention

[0004] The purpose of the present invention is to provide an API gateway recommendation method and system based on multimodal feature fusion to solve the following technical problems:

[0005] How to coordinate and integrate multimodal data features with API gateway architecture information to promote the optimization of API gateway recommendation process.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A multimodal feature fusion API gateway recommendation method, the method comprising:

[0008] Step 1: The API gateway receives the client service request authentication at the entrance, routes the multimodal data through the multi-protocol proxy module, and requests the filtering plug-in for preliminary verification;

[0009] Step 2: The receiving layer receives the image processing service request and the text processing service request, extracts features from the multimodal data after preliminary verification, and stores the features;

[0010] Step 3: The feature layer receives the call information, fuses the features of the same processing node, confirms the fusion strategy and sends it to the decision layer;

[0011] Step 4: The decision-making layer generates API gateway recommendations based on the fusion strategy and generates recommendation engine feedback output to the client.

[0012] Preferably, the multimodal data is routed through the multi-protocol proxy module and the filtering plug-in is requested to perform preliminary verification as follows:

[0013] S1. The API gateway server obtains the port of the listening protocol and receives the client service request;

[0014] S2. Parse multimodal data according to protocol type and extract structured and unstructured data;

[0015] S3: Inject protocol type features into the client service request context, route and filter structured and unstructured data, and use filtering plug-ins to filter and clean structured and unstructured data.

[0016] S4. Convert the routed and filtered structured data and unstructured data into API internal standard format data.

[0017] Preferably, the feature extraction method in step 2 is:

[0018] SS1. Obtain the API gateway server interface information selected by the client, create an API service request for the client based on the API key selected by the client, and send the API service request to the API gateway server;

[0019] SS2, the API gateway server receives multiple service request calls from the client and marks them, and analyzes the request response time and request response data type of each service request call;

[0020] SS3. Obtain the similarity of service request calls with the same request response time based on the request response time and request response data type:

[0021] When the similarity of service request calls under the same request response time is greater than the target similarity, it is determined to be a similar service request call, and the service progress information of all similar service request calls is synchronized;

[0022] Otherwise, it is determined that the calls are different service requests, and service progress information of the different service request calls is obtained respectively;

[0023] SS4 and the API gateway server call the same service progress information, extract the service request features according to the historical feature parsing protocol mapping table, and obtain the service request features.

[0024] Preferably, the types of service request features include image processing service request features and text processing service request features;

[0025] When requesting characteristics for the Image Processing service:

[0026] Input the image tensor in standard format into the ResNe model and output the image vector;

[0027] When requesting features for the Text Processing service:

[0028] Input the standard format sentence segmentation into the BERT model and output the sentence vector;

[0029] When both image processing service request features and text processing service request features are included:

[0030] The API sends a call to perform feature fusion processing on the standard format data.

[0031] Preferably, the fusion strategy is:

[0032] Acquire feature vector information of the same processing node, and confirm the feature cache flow of the graph vector and the feature cache flow of the sentence vector in the feature vector information;

[0033] Calculate the feature cache flow mean of the graph vector and the feature cache flow mean of the sentence vector for the same processing node;

[0034] The recommendation level of the client request response path is determined by calculating the ratio of the feature cache flow mean of the graph vector and the feature cache flow mean of the sentence vector.

[0035] By judging the ratio and the preset ratio threshold:

[0036] If the ratio falls within the preset ratio threshold, the recommended level of the client request response path remains unchanged at the current level;

[0037] If the ratio is greater than the preset ratio threshold, the recommendation level of the client request response path increases by one level;

[0038] If the ratio is less than the preset ratio threshold, the recommendation level of the client request response path is reduced by one level;

[0039] A CLIP model mapping is established to calculate the weighted average of the graph vector and the sentence vector, and the dynamic matching degree of the client request response path is output based on the weighted average size and recommendation level.

[0040] Preferably, the recommended steps for the API gateway are:

[0041] Determine whether the dynamic matching degree meets the target value set by the response API gateway:

[0042] If yes, the user profile and weighted average value loaded from the database are used as input to generate a recommendation engine;

[0043] If not, an error message is generated and the client service request authentication is returned.

[0044] Preferably, the service request calling process of the API gateway is:

[0045] The client creates and initiates an API service call request, and sends the created API service call request to the API gateway server;

[0046] The API gateway server processes the API service call request API and sends the response request data back to the client;

[0047] The client parses the API response request data and extracts recommendation information from it and feeds it back to the client.

[0048] An API gateway recommendation system with multimodal feature fusion, including:

[0049] The verification module is used to receive client service request authentication through the API gateway entrance, route the multimodal data through the multi-protocol proxy module, and request the filtering plug-in for preliminary verification;

[0050] The extraction and storage module is used for receiving the image processing service request and the text processing service request at the receiving layer, extracting features from the multimodal data after preliminary verification, and storing the features;

[0051] The fusion module is used to receive call information from the feature layer, fuse the features of the same processing node, confirm the fusion strategy and send it to the decision layer;

[0052] The recommendation module is used to generate API gateway recommendations based on the fusion strategy of the decision layer and generate recommendation engine feedback output to the client.

[0053] Beneficial effects of the present invention:

[0054] (1) The design of the preliminary verification step is to match the type of multimodal data and API protocol to achieve the parsing of multimodal data and extract the data structure content that conforms to the API gateway. The structured data and unstructured data after filtering are converted into API internal standard format data according to the protocol type; the subsequent authentication and balance confirmation process of multimodal data feature extraction through the API receiving layer is omitted.

[0055] (2) The client service request authentication is confirmed through the receiving layer based on the API gateway, and the image processing service request and text processing service request are confirmed by responding to the different service request call information of the client in the server, and the feature extraction and feature storage process of the multimodal data after the first step of preliminary verification is performed; the accurate extraction process of the service request feature is ensured, and the response method recommended by the API gateway is quickly screened.

[0056] (3) The feature layer receives the call information to confirm the fusion strategy and sends it to the decision layer, and the features of the same processing node are fused to ensure the fast and efficient fusion of feature information. A fusion strategy is designed to obtain feature vector information at the same processing node and analyze the data cache traffic of multimodal feature data to match the results. The recommended path and recommendation method are confirmed based on the data vectors of different modes fed back by the client. The recommendation level is determined by the current API gateway usage scenario and user cluster calculation. The higher the recommendation level, the faster the response request; the higher the allocated traffic; the coordinated fusion of multimodal data features and API gateway architecture information is achieved, and the optimization of the API gateway recommendation process is promoted.

[0057] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0059] Figure 1 This is a step diagram of a method for recommending an API gateway for multimodal feature fusion according to the present invention;

[0060] Figure 2 This is a process diagram of the present invention, wherein multimodal data is routed through a multi-protocol proxy module and a filter plug-in is requested to perform preliminary verification.

[0061] Figure 3 This is a diagram of the feature extraction steps of the present invention;

[0062] Figure 4 This is a flowchart of the service request calling method of the API gateway of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] See also Figure 1 As shown, the present invention is an API gateway recommendation method based on multimodal feature fusion, which specifically includes:

[0065] Step 1: The API gateway receives the client service request authentication at the entrance, routes the multimodal data through the multi-protocol proxy module, and requests the filtering plug-in for preliminary verification;

[0066] Step 2: The receiving layer receives the image processing service request and the text processing service request, extracts features from the multimodal data after preliminary verification, and stores the features;

[0067] Step 3: The feature layer receives the call information, fuses the features of the same processing node, confirms the fusion strategy and sends it to the decision layer;

[0068] Step 4: The decision-making layer generates API gateway recommendations based on the fusion strategy and generates recommendation engine feedback output to the client.

[0069] In the above technical solutions, in the design of the recommendation system, it is usually necessary to build an API recommendation framework and design the request recommendation service at the appropriate location in the API framework. First, the request is routed through the gateway, and the multimodal data is distributed to different layers (receiving layer, feature layer, decision layer) for processing services. The gateway's load balancing is used to ensure stability under high concurrency. Choosing to fuse the features of multimodal data at the API's feature layer and to fuse multimodal data through the decision layer usually affects the API gateway recommendation service design process. In addition, the feature layer fusion considers the API gateway's routing of data of different modalities to the same processing node, while the decision layer fusion considers the efficiency of aggregating the output results of multiple services. At the same time, the API gateway also requires a cache mechanism that can be used to store commonly used recommendation results to improve response speed.

[0070] Based on the above considerations, we designed an API gateway recommendation method based on multimodal feature fusion, which integrates multimodal data features with API gateway architecture information to promote the design process of API gateway recommendation. The specific steps are as follows:

[0071] First, the first step is to select the entry protocol of the API gateway, and realize the unified reception of heterogeneous data formats through protocol conversion capabilities (such as HTTP / HTTPS to gRPC); the entry protocol is used to receive data sources of different modalities and route them to the designated location through the gateway. The specific method is to receive the client service request authentication at the API gateway entry, access the multimodal data through the multi-protocol proxy module set by the API gateway, and route it to the request filtering plug-in for preliminary verification; this embodiment realizes the parsing of multimodal data by matching the type of multimodal data and API protocol through the design of the preliminary verification step, and extracts the data structure content that conforms to the API gateway, eliminating the subsequent authentication and balance confirmation process of multimodal data feature extraction through the API receiving layer, and the API gateway realizes preliminary data verification through the request filtering plug-in, such as detecting whether the image resolution meets the standard and whether the text contains sensitive words, and intercepts abnormal requests at the entry layer, thereby reducing the operating load of other API layers.

[0072] The specific preliminary verification process is as follows:

[0073] As an embodiment of the present invention, please refer to Figure 2 As shown in the figure, the process of routing multimodal data through the multi-protocol proxy module and requesting the filtering plug-in to perform preliminary verification is as follows:

[0074] S1. The API gateway server obtains the port of the listening protocol and receives the client service request;

[0075] S2. Parse multimodal data according to protocol type and extract structured and unstructured data;

[0076] S3: Inject protocol type features into the client service request context, route and filter structured and unstructured data, and use filtering plug-ins to filter and clean structured and unstructured data.

[0077] S4. Convert the routed and filtered structured data and unstructured data into API internal standard format data.

[0078] In the above technical solution, the API gateway server first obtains the port of the listening protocol, and the listening protocol is determined according to the user agreement; it is convenient to receive the client's service request in real time; then, before accessing the route, the client service request at the port is received through the multi-protocol proxy module, and the multimodal data is parsed according to the protocol type to extract the structured and unstructured content; then the protocol type feature is injected into the context of the client service request, and the extracted structured and unstructured content is routed. During the routing process, the structured data and unstructured data are filtered through the filtering plug-in, and the information that does not meet the filtering conditions is cleaned and intercepted; finally, the structured data and unstructured data after filtering are converted into API internal standard format data according to the protocol type.

[0079] Then, the second step is to receive the client service request authentication through the receiving layer of the API gateway architecture, confirm the image processing service request and the text processing service request, and extract and store the features of the multimodal data after the preliminary verification in the first step; the feature extraction in the second step is different from the conventional multimodal structure data extraction, but is to confirm the client service request authentication through the receiving layer based on the API gateway, and through the confirmation and processing method of the different service request call information of the customer in the response server, to ensure the accurate extraction process of the service request feature, and quickly screen the response method recommended by the API gateway; the client service request authentication includes before and after conversion into internal standard format data, and preliminarily determines whether the data requested by the customer has passed the basic security verification, client identity authentication and permission verification.

[0080] Specifically, as an embodiment of the present invention, please refer to Figure 3 As shown, the feature extraction method in step 2 is:

[0081] SS1. Obtain the API gateway server interface information selected by the client, create an API service request for the client based on the API key selected by the client, and send the API service request to the API gateway server;

[0082] SS2, the API gateway server receives multiple service request calls from the client and marks them, and analyzes the request response time and request response data type of each service request call;

[0083] SS3. Obtain the similarity of service request calls with the same request response time based on the request response time and request response data type:

[0084] When the similarity of service request calls under the same request response time is greater than the target similarity, it is determined to be a similar service request call, and the service progress information of all similar service request calls is synchronized;

[0085] Otherwise, it is determined that the calls are different service requests, and service progress information of the different service request calls is obtained respectively;

[0086] SS4 and the API gateway server call the same service progress information, extract the service request features according to the historical feature parsing protocol mapping table, and obtain the service request features.

[0087] In the above technical solution, first, a specific service request is confirmed for the client. Specifically, the API gateway server interface information selected by the client is obtained, and the API key selected by the client is obtained, an API service request is created for the client, and the API service request is sent to the API gateway server; then, after the client request service authentication is passed, the API gateway server receives and marks multiple service request calls from the client, and the response method recommended by the API gateway can be accurately and clearly confirmed based on the service request call, and the response method recommended by the API gateway is confirmed by analyzing the request response time and request response data type of each service request call.

[0088] Unlike conventional feature extraction of only structured data and unstructured data, it is necessary to further screen the degree of client request calls based on the request response time and request response data type, specifically by obtaining the similarity of service request calls under the same request response time; the similarity information of the service request call is based on the client's behavior sequence and the "client-service request call" matrix decomposition in the large database, and the time series of the real-time client behavior sequence is registered to the API gateway server. The API gateway server confirms the normalized time series features and builds the client's time series analysis service based on the client's time series features to confirm the similarity of different service request calls; for example, define "service request call" as "product browsing", The historical "client-product browsing" matrix decomposition in the big database contains the number of times the client has browsed the product and the duration of their stay. Based on the client's frequent behavior in the historical time period (more than i views, for example, 3 times; or a browsing time between 2 and 5 minutes, it is recorded as 1, otherwise it is 0), a time series feature statistics for the time period is constructed, such as {1, 0, 1..., 0, 1, 1}. The number of time series features with a value of 1 in the time period is counted as the value of the service request call. When the difference is obtained from the number of views with a value of 1 in the historical "client-product" matrix decomposition, the ratio of the difference to the total number of views is defined as the similarity of the product browsing.

[0089] The method for judging similarity is:

[0090] Specifically, the confirmation of the same or similar service request calls issued by the client is determined based on the similarity calculation of the service request calls, and then the service request is further synchronously responded to. The specific judgment method is: when the similarity of the service request calls under the same request response time is greater than the target similarity, it is determined to be a similar service request call, and the service progress information of all similar service request calls is synchronized; otherwise, it is determined to be a different service request call, and the service progress information of the different service request calls is obtained respectively; the service progress is specifically in the product browsing project to push the product category to the client, the client confirms receipt, and the client feedback recommendation stage program is displayed with service progress. Finally, the API gateway server calls the same service progress information, extracts service request features for a type of client service request according to the historical feature analysis protocol mapping table, obtains service request features, and forms a type of service request progress information that is output from the receiving layer to the feature layer.

[0091] Then, in the third step, the feature layer receives the call information and fuses the features of the same processing node, confirms the fusion strategy and sends it to the decision layer; the fusion strategy in step three ensures the fast and efficient fusion of feature information.

[0092] As an implementation of the present invention, the specific fusion strategy is:

[0093] Obtain feature vector information for the same processing node and confirm the feature cache flow of the graph vector and the sentence vector in the feature vector information. Confirming the feature cache flow at the same processing node ensures that during feature layer fusion, the API gateway routes data of different modalities to the same processing node, resulting in a large amount of data cache flow. Calculate the mean feature cache flow of the graph vector and the sentence vector for the same processing node. Confirm the recommended path and method based on the data vectors of different modalities fed back by the client.

[0094] The specific recommendation method is to determine the recommendation level of the client request response path by calculating the ratio of the feature cache traffic mean of the graph vector and the feature cache traffic mean of the sentence vector; by judging the size of the ratio and the preset ratio threshold, if the ratio belongs to the preset ratio threshold, the recommendation level of the client request response path remains unchanged at the current level; if the ratio is greater than the preset ratio threshold, the recommendation level of the client request response path is increased by one level; if the ratio is less than the preset ratio threshold, the recommendation level of the client request response path is reduced by one level; the recommendation level is determined according to the current API gateway usage scenario and user cluster calculation to determine the grading system. The higher the recommendation level, the faster the response request; and the higher the allocated traffic.

[0095] The CLIP model mapping is also established to calculate the weighted average of the graph vector and the sentence vector. The dynamic matching degree of the client request response path is output based on the weighted average value and the recommendation level. For example, assume that the recommendation level R is marked as "1", indicating "highly recommended"; R is marked as "0", indicating "generally recommended"; and R is marked as "-1", indicating "not recommended". The CLIP model mapping calculation obtains the CLIP similarity of the graph vector and the sentence vector respectively, and uses cosine similarity to represent them as C1 and C2 respectively. The proportion of the feature cache traffic of the historical graph vector and sentence vector in the total cache traffic, W1 and W2, is obtained respectively, reflecting the weight of the historical graph vector and sentence vector. The process of calculating the weighted average S is as follows: ; Dynamic matching calculation (R is not equal to 0) is: ,in is the grade coefficient, usually 0.1; when R is "-1" ; When R is "1", ; When R is "0", .

[0096] Finally, the fourth step is for the decision-making layer to generate API gateway recommendations based on the fusion strategy, generate recommendation engine feedback and output it to the client, so that the API gateway can make recommendation feedback on the results of the fusion strategy.

[0097] Specifically, as an embodiment of the present invention, the API gateway recommends the following steps:

[0098] Determine whether the dynamic matching degree meets the target value set by the response API gateway:

[0099] If yes, the user profile and weighted average value loaded from the database are used as input to generate a recommendation engine;

[0100] If not, an error message is generated and the client service request authentication is returned.

[0101] In the above technical solution, since the decision-layer fusion needs to consider the efficiency of aggregating the output results of multiple services, in this embodiment, a target value of the API gateway matching degree is set (obtained by machine training), and the dynamic matching degree of the feature layer fusion output is judged to determine whether the target value meets the target value set by the response API gateway: if so, the user portrait loaded from the database and the weighted average are used as input to generate a recommendation engine; if not, an error message is generated and the client service request authentication is returned.

[0102] As an embodiment of the present invention, please refer to Figure 4 As shown, the overall process of API gateway service request call is:

[0103] The client creates and initiates an API service call request, and sends the created API service call request to the API gateway server;

[0104] The API gateway server processes the API service call request API and sends the response request data back to the client;

[0105] The client parses the API response request data and extracts recommendation information from it and feeds it back to the client.

[0106] As an embodiment of the present invention, the types of service request features include image processing service request features and text processing service request features;

[0107] When requesting characteristics for the Image Processing service:

[0108] Input the image tensor in standard format into the ResNe model and output the image vector;

[0109] When requesting features for the Text Processing service:

[0110] Input the standard format sentence segmentation into the BERT model and output the sentence vector;

[0111] When both image processing service request features and text processing service request features are included:

[0112] The API sends a call to perform feature fusion processing on the standard format data.

[0113] The embodiment of the present invention also includes an API gateway recommendation system for multimodal feature fusion, which is provided with:

[0114] The verification module is used to receive client service request authentication through the API gateway entrance, route the multimodal data through the multi-protocol proxy module, and request the filtering plug-in for preliminary verification;

[0115] The extraction and storage module is used for receiving the image processing service request and the text processing service request at the receiving layer, extracting features from the multimodal data after preliminary verification, and storing the features;

[0116] The fusion module is used to receive call information from the feature layer, fuse the features of the same processing node, confirm the fusion strategy and send it to the decision layer;

[0117] The recommendation module is used to generate API gateway recommendations based on the fusion strategy of the decision layer and generate recommendation engine feedback output to the client.

[0118] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.

[0119] The above content is merely an example and explanation of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they should all fall within the scope of protection of the present invention.

Claims

1. A multimodal feature fusion API gateway recommendation method, characterized in that: The method comprises: Step 1: The API gateway receives the client service request authentication at the entrance, routes the multimodal data through the multi-protocol proxy module, and requests the filtering plug-in for preliminary verification; Step 2: The receiving layer receives the image processing service request and the text processing service request, extracts features from the preliminarily verified multimodal data, and stores the features. The feature extraction method in step 2 is: SS1. Obtain the API gateway server interface information selected by the client, create an API service request for the client based on the API key selected by the client, and send the API service request to the API gateway server; SS2, the API gateway server receives multiple service request calls from the client and marks them, and analyzes the request response time and request response data type of each service request call; SS3. Obtain the similarity of service request calls with the same request response time based on the request response time and request response data type: When the similarity of service request calls under the same request response time is greater than the target similarity, it is determined to be a similar service request call, and the service progress information of all similar service request calls is synchronized; Otherwise, it is determined that the calls are different service requests, and service progress information of the different service request calls is obtained respectively; SS4, the API gateway server calls the same service progress information, extracts the service request features according to the historical feature parsing protocol mapping table, and obtains the service request features; Step 3: The feature layer receives the call information, fuses the features of the same processing node, and confirms the fusion strategy and sends it to the decision layer; the fusion strategy is: Acquire feature vector information of the same processing node, and confirm the feature cache flow of the graph vector and the feature cache flow of the sentence vector in the feature vector information; Calculating a feature cache flow mean of the graph vector and a feature cache flow mean of the sentence vector of the same processing node; The recommendation level of the client request response path is determined by calculating the ratio of the feature cache flow mean of the graph vector and the feature cache flow mean of the sentence vector. By judging the ratio and the preset ratio threshold: If the ratio falls within the preset ratio threshold, the recommended level of the client request response path remains unchanged at the current level; If the ratio is greater than the preset ratio threshold, the recommendation level of the client request response path increases by one level; If the ratio is less than the preset ratio threshold, the recommendation level of the client request response path is reduced by one level; Establish a CLIP model mapping to calculate the weighted average of the graph vector and sentence vector, and output the dynamic matching degree of the client request response path based on the weighted average value and recommendation level; Step 4: The decision-making layer generates API gateway recommendations based on the fusion strategy, generates recommendation engine feedback and outputs it to the client. The API gateway recommendation steps are as follows: Determine whether the dynamic matching degree meets the target value set by the response API gateway: If yes, the user profile and weighted average value loaded from the database are used as input to generate a recommendation engine; If not, an error message is generated and the client service request authentication is returned.

2. The API gateway recommendation method based on multimodal feature fusion according to claim 1 is characterized in that: The process of accessing the multimodal data to the router through the multi-protocol proxy module and requesting the filtering plug-in to perform preliminary verification is as follows: S1. The API gateway server obtains the port of the listening protocol and receives the client service request; S2. Parse multimodal data according to protocol type and extract structured and unstructured data; S3: Inject protocol type features into the client service request context, route and filter structured and unstructured data, and use filtering plug-ins to filter and clean structured and unstructured data. S4. Convert the routed and filtered structured data and unstructured data into API internal standard format data.

3. The API gateway recommendation method based on multimodal feature fusion according to claim 1 is characterized in that: The types of service request features include image processing service request features and text processing service request features: When requesting characteristics for the Image Processing service: Input the image tensor in standard format into the ResNe model and output the image vector; When requesting features for the Text Processing service: Input the standard format sentence segmentation into the BERT model and output the sentence vector; When both image processing service request features and text processing service request features are included: The API sends a call to perform feature fusion processing on the standard format data.

4. The API gateway recommendation method based on multimodal feature fusion according to claim 1 is characterized in that: The service request calling process of the API gateway is: The client creates and initiates an API service call request, and sends the created API service call request to the API gateway server; The API gateway server processes the API service call request and sends the API response request data back to the client; The client parses the API response request data and extracts recommendation information from it and feeds it back to the client.

5. A multimodal feature fusion API gateway recommendation system, which adopts the multimodal feature fusion API gateway recommendation method according to any one of claims 1 to 4, characterized in that: include: The verification module is used to receive client service request authentication through the API gateway entrance, route the multimodal data through the multi-protocol proxy module, and request the filtering plug-in for preliminary verification; The extraction and storage module is used to receive image processing service requests and text processing service requests from the receiving layer, extract features from the multimodal data after preliminary verification, and store the features; the feature extraction method is: SS1. Obtain the API gateway server interface information selected by the client, create an API service request for the client based on the API key selected by the client, and send the API service request to the API gateway server; SS2, the API gateway server receives multiple service request calls from the client and marks them, and analyzes the request response time and request response data type of each service request call; SS3. Obtain the similarity of service request calls with the same request response time based on the request response time and request response data type: When the similarity of service request calls under the same request response time is greater than the target similarity, it is determined to be a similar service request call, and the service progress information of all similar service request calls is synchronized; Otherwise, it is determined that the calls are different service requests, and service progress information of the different service request calls is obtained respectively; SS4, the API gateway server calls the same service progress information, extracts the service request features according to the historical feature parsing protocol mapping table, and obtains the service request features; The fusion module is used to receive call information from the feature layer, fuse the features of the same processing node, confirm the fusion strategy and send it to the decision layer; the fusion strategy is: Acquire feature vector information of the same processing node, and confirm the feature cache flow of the graph vector and the feature cache flow of the sentence vector in the feature vector information respectively; Calculating a feature cache flow mean of the graph vector and a feature cache flow mean of the sentence vector of the same processing node; The recommendation level of the client request response path is determined by calculating the ratio of the feature cache flow mean of the graph vector and the feature cache flow mean of the sentence vector. By judging the ratio and the preset ratio threshold: If the ratio falls within the preset ratio threshold, the recommended level of the client request response path remains unchanged at the current level; If the ratio is greater than the preset ratio threshold, the recommendation level of the client request response path increases by one level; If the ratio is less than the preset ratio threshold, the recommendation level of the client request response path is reduced by one level; Establish a CLIP model mapping to calculate the weighted average of the graph vector and sentence vector, and output the dynamic matching degree of the client request response path based on the weighted average value and recommendation level; The recommendation module is used to generate API gateway recommendations based on the fusion strategy of the decision-making layer and generate recommendation engine feedback output to the client; the API gateway recommendation steps are: Determine whether the dynamic matching degree meets the target value set by the response API gateway: If yes, the user profile and weighted average value loaded from the database are used as input to generate a recommendation engine; If not, an error message is generated and the client service request authentication is returned.

Images