Intelligent distributed interaction method for airborne environment

Abstract

The invention discloses an airborne environment-oriented intelligent distributed interaction method, which comprises the following steps that: a data publisher classifies related data obtained by the data publisher according to themes, and uploads the classified data to a theme message queue of a server for a user to obtain; the related data comprises message data and environment data; and the user obtains data from the theme message queue according to own requirements. According to the method and the device, required information can be accurately obtained from mass message data in an airborne environment, and a data access mode of a user not only can support an active data query demand, but also has an intelligent data pushing capability.

Description

Technical Field

[0001] This application relates to the field of airborne technologies, and particularly to an intelligent distributed interaction method for an airborne environment. Background Art

[0002] In the modern context, the complexity and diversity of the airborne environment require that various units or nodes can quickly and accurately share and process a large amount of data information to complete predetermined tasks. These data not only include traditional messages and information, but may also involve various aspects of information such as meteorological data and geographical locations. However, in the face of a large amount of environmental data and message information in the airborne environment, the traditional request-response mode of information interaction, that is, the client initiates a request, and the server responds to the request and returns data. In the case of a large amount of data and high network latency, there are problems such as long response time and poor scalability in data interaction. Using a single request-response mode for data sharing can no longer meet the requirements of airborne data interaction. Therefore, there is an urgent need to provide a more proactive data acquisition method for users.

[0003] To improve the interaction efficiency and response speed, currently, distributed interaction technologies adopted in the Internet environment include publish / subscribe, message queue, long polling, etc. However, in a complex airborne environment, the number of role types of each unit is small, and there are also certain differences in data access behaviors within the roles. Traditional recommendation algorithms applied to the Internet cannot be directly applied to the airborne environment. To meet the requirements of the rapidly changing airborne environment, improve efficiency, accelerate the response speed, enhance flexibility, ensure the accuracy of command and control, and provide users with a more proactive and intelligent data service, after building a basic data storage and management platform at the bottom layer, the intelligent distributed interaction technology at the application layer has become the research focus of data interaction in the airborne environment. Summary of the Invention

[0004] To solve the problem that it is difficult for each unit in the airborne environment to accurately obtain the required information from a large amount of message data through the traditional request-response data acquisition method, this application provides an intelligent distributed interaction method for an airborne environment, enabling the user's data access mode to support both proactive data query requirements and have a certain intelligent data push ability.

[0005] This application discloses an intelligent distributed interaction method for an airborne environment, which includes: The data publisher classifies the relevant data it obtains according to the topic, and uploads the classified data to the topic message queue of the server for users to obtain; the relevant data includes message data and environmental data; the user is an airborne platform; The user obtains data from the topic message queue according to their own needs.

[0006] Further, the user obtains data from the topic message queue of the server according to his own needs, including: For topics that the user has subscribed to before, the user directly obtains data from the topic message queue through the publish / subscribe management function of the server without going through the recommendation of the server's intelligent recommendation mechanism; For topics that the user has not subscribed to before, the server's data topic push function is used to establish an association between each user and each topic. Based on the intelligent recommendation mechanism, the degree of interest of each user in each topic is measured. According to the user's historical subscription records, historical access records, and user feedback scores, user preference feature vectors are extracted. The similarity between the user preference feature vector of each user and the topic feature vector of each data topic is calculated, and this similarity is used as a recommendation factor; the recommendation factor is used to provide a recommendation function for the user in addition to the data publish / subscribe service.

[0007] Further, the providing of the recommendation function for the user in addition to the data publish / subscribe service by the recommendation factor includes: When the recommendation factor is greater than or equal to the recommendation threshold, it indicates that the recommended topic is valuable to the user. The recommended topic is added to the recommendation list, and a recommendation message is sent to the user; when the recommendation factor is less than the recommendation threshold, the intelligent recommendation mechanism ignores the recommended topic and waits to re-judge whether to recommend the topic when the user's preferences change.

[0008] Further, after the user receives the recommended topic, if the recommended topic is a topic of interest to the user, a request is initiated to the server's data topic push function. The server's data topic push function actively obtains the corresponding message data from the topic message queue. The access information of the user to the server's data topic push function, that is, the historical access record, will be fed back to the user preferences to affect the portrayal of the user profile; when the recommended topic is not a topic of interest to the user, the user's feedback will also affect the next portrayal of the user profile.

[0009] Further, the intelligent recommendation mechanism includes: Topic representation: Extract the features of the topics in the topic message queue through the TF-IDF keyword extraction technology, which is the topic feature vector, and use this topic feature vector to represent the recommended topic; Feature learning: According to the user's historical access and subscribed topic records, and the feedback scores on the recommended topics, obtain the task preference vector through the LDA algorithm, and obtain the recommended benchmark vector through the collaborative filtering recommendation mechanism. Combine the task preference vector and the recommended benchmark vector to portray the user profile to obtain the user preference feature vector; Generate a recommendation list: Calculate the similarity between the topic feature vector and the user preference feature vector through the cosine similarity function, use the similarity as a recommendation factor, and generate a recommendation list based on the recommendation factor and the recommendation threshold.

[0010] Further, the features of the topics in the topic message queue are extracted through the TF-IDF keyword extraction technology, which are the topic feature vectors, including: The topic in the topic message queue is a set containing N topic message data, that is, { }, where the number of message data containing the keyword is pieces, represents the number of times the keyword appears in the message data , then The word frequency in the message data is defined as the number of times appears in the total number of occurrences of all z keywords in the message data , that is:

[0011] Among them, z represents the number of keywords that appear in the message data , that is, there are keywords { } in the message data , represents the keyword appears in the message data , represents the sum of the number of occurrences of all keywords in the message data; The inverse frequency that appears in the topic is defined as:

[0012] Among them, represents the total number of messages in the topic, represents the number of files containing the keyword ; The TF-IDF parameter is defined as:

[0013] Use a k-dimensional vector to represent the features of the topics in the topic message queue.

[0014] Further, in the feature learning: Obtain the user's historical access records and historical subscription records in recent tasks, and combine the feedback scoring results of the user on the intelligent recommendation topics; extract the set of the user's interest topics from the historical records through the LDA algorithm, and construct it into a task preference vector; The recommended benchmark vector obtained through the collaborative filtering recommendation mechanism; Use the task preference vector as the main part for calculating the user preference feature vector, and use the deviation between the task preference vector of each user and the recommended benchmark vector as the secondary part for calculating the user preference feature vector to obtain the user preference feature vector.

[0015] Further, the recommended benchmark vector obtained through the collaborative filtering recommendation mechanism includes: Calculate similarity: Based on the task preference vector of each user, calculate the similarity between the target user and other users through cosine similarity; Identify similar user groups: Determine a similarity threshold or select the top N approximate users as the similar user groups; Determine the recommended topics: Collect the set of interest topics of the target user's similar user groups, including the subscribed and accessed topics, to form the recommended benchmark vector.

[0016] Further, the update method of the recommended factor includes: Input: User preference feature P, feature learning algorithm A, all topic information S; 1) : Extract the topic feature vector from the topic information through the TF-IDF keyword extraction method; is the topic feature vector; is the extraction function; 2) Compare the user preference feature P with the topic feature vector IR_S, use the VSM model, calculate the similarity between P and IR_S, sort the similarities, that is, sort the recommended factors, to obtain the recommended list R1; 3) The user scores the recommended information in the recommended list R1, there are explicit feedback and implicit feedback, and the feedback information is t; the explicit feedback is the scoring information; the implicit feedback is the historical access information; 4) Update the user preference feature , is the update function, t is the time, to improve the recommendation accuracy; 5) Return to step 2), continue to generate the recommended list R1; Output: The user's recommended list R1, the user preference feature P.

[0017] Further, verify the effectiveness of the recommended factor through relevant indicators; the relevant indicators include coverage rate, discoverability and score; Coverage rate: It represents the ability of the intelligent recommendation mechanism to discover message topics. The coverage rate is defined as follows:

[0018] where N represents the total number of user message topics, represents the list of topics recommended by the intelligent recommendation mechanism to the user, represents the number of topics included in the recommended topic list, represents the topic in the topic list recommended to the user; Discoverability: It represents the ability of the intelligent recommendation mechanism to recommend to the user the message data that the user has not subscribed to but is suitable for the user's topic. The discoverability is defined by the following expression:

[0019] where n is the number of historical recommendations counted, is the number of topics that were not selected in the previous historical selections of a certain recommended statistics but were selected this time, is the number of topics that were selected in this recommended and were also selected before; Score: It represents the degree of satisfaction of the user with the topic list recommended by the intelligent recommendation mechanism. The score is evaluated using precision and recall. Precision is defined as:

[0020] Recall is defined as:

[0021] where, is the precision, is the recall, R(u) represents the topic list recommended by the intelligent recommendation mechanism to the user, represents the topic list that the user really needs, is all the topics in the topic message queue.

[0022] Due to the adoption of the above technical solution, the present application has the following advantages: 1. This application adopts a message subscription / publishing architecture design as the user data acquisition method, constructs an abstraction layer between the data publisher and the user. Both communication parties only need to focus on their own message data and task execution, etc., without having to pay attention to the network transmission process. The subscription / publishing architecture avoids the defects of traditional periodic polling technology in real-time communication scenarios in a complex airborne environment, and achieves a balance between bandwidth occupancy and communication real-time performance. Whether it is the data publisher or the user, they both generate and consume data in a topic-based manner, rather than through network addresses. Therefore, the topology structure in the network becomes simpler, and even when both communication parties dynamically join and leave, they can adapt relatively quickly. The topic-based subscription / publishing architecture can support QoS configuration according to resource availability, the degree of resource occupancy by the provider, the expectation degree of the requester for resources, etc., to improve the flexibility of network communication.

[0023] 2. This application adopts a content-based recommendation algorithm and a collaborative filtering recommendation mechanism as the intelligent recommendation mechanism, which can recommend topics of interest to users from a large amount of data. This service of actively pushing message topics to users can reduce unnecessary data transmission in the data link, reduce bandwidth occupancy, and increase efficiency. And after accumulating a large amount of data in the later stage of the recommendation system, the model can be upgraded to a hybrid algorithm and a deep learning algorithm to improve the recommendation quality in the later stage.

[0024] 3. This application combines the data subscription / publishing architecture design with a push / pull data acquisition method based on topic recommendation factors to solve the difficult problem of each user within the user obtaining useful information from a large amount of message data and information, making the data interaction method based on the subscription / publishing architecture design more applicable to the airborne environment, effectively improving efficiency and performance, and realizing a distributed data interaction function based on the data subscription / publishing mechanism, message queue and intelligent recommendation mechanism, including data push mechanism, subscription management mechanism, topic message queue and intelligent recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings.

[0026] Figure 1 It is the data publishing / subscribing architecture of an intelligent distributed interaction method for an airborne environment in an embodiment of the present application.

[0027] Figure 2 It is the intelligent push working principle of an intelligent distributed interaction method for an airborne environment in an embodiment of the present application. Detailed implementation manners

[0028] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only a part of the embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art shall fall within the scope of protection of the embodiments of the present application.

[0029] See Figure 1 , an embodiment of an intelligent distributed interaction method for an airborne environment provided by the present application includes: The data publisher classifies the relevant data it obtains according to the topic, and uploads the classified data to the topic message queue of the server for users to obtain; the relevant data includes message data and environmental data; the user is an airborne platform; The user obtains data from the topic message queue according to his own needs.

[0030] See Figure 1 , the server in the embodiment of the present application has a publish / subscribe management function, a data topic push function, and a stored topic message queue and intelligent recommendation mechanism.

[0031] In this embodiment, the setting of the data topic is based on the principle of ensuring that the message system is both flexible and reliable, clarifying the purpose of the topic and the type of message to be transmitted, and the message topic can be set according to business requirements, message format, message size and frequency, message persistence, etc.

[0032] In this embodiment, the topic message queue is used for efficient message transmission between the data publisher and the user. The publisher stores the data related to the topic in the topic message queue until the user consumes the data, realizing decoupling between the publisher and the user.

[0033] Optionally, the user obtains data from the topic message queue of the server according to his own needs, including: For the topics that the user has subscribed to before, the user directly obtains data from the topic message queue through the publish / subscribe management function of the server without going through the recommendation of the intelligent recommendation mechanism of the server; For the topics that the user has not subscribed to before, the data topic push function of the server establishes an association between each user and each topic. Based on the intelligent recommendation mechanism, the degree of interest of each user in each topic is measured, and the user preference feature vector is extracted according to the user's historical subscription records and historical access records. The similarity between the user preference feature vector of each user and the topic feature vector of each data topic is calculated, and this similarity is used as the recommendation factor; through the recommendation factor, a recommendation function other than the data publish / subscribe service is provided for the user, thereby improving the efficiency and intelligence of the data interaction mechanism.

[0034] Optionally, provide a recommendation function for the user in addition to the data publishing / subscribing service through a recommendation factor, including: When the recommendation factor is greater than or equal to the recommendation threshold, it indicates that the recommended topic is valuable to the user. Add the recommended topic to the recommendation list and send a recommendation message to the user; when the recommendation factor is less than the recommendation threshold, the intelligent recommendation mechanism ignores the recommended topic and waits to re-judge whether to recommend the topic when the user's preference changes.

[0035] Optionally, after the user receives the recommended topic, if the recommended topic is a topic of interest to the user, initiate a request through the data topic push function of the server. The data topic push function of the server actively obtains the corresponding message data from the topic message queue. The access information of the user to the data topic push function of the server, that is, the historical access record, will be fed back to the user preference to affect the portrayal of the user profile; when the recommended topic is not a topic of interest to the user, the user's feedback will also affect the next portrayal of the user profile.

[0036] Optionally, the intelligent recommendation mechanism includes: Topic representation: Extract the features of the topic in the topic message queue through the TF-IDF (Term Frequency-Inverse Document Frequency) keyword extraction technology, which is the topic feature vector, and represent the recommended topic with the topic feature vector; Feature learning: According to the user's historical access and subscribed topic records, as well as the feedback score on the recommended topic, obtain the task preference vector through the LDA algorithm, and obtain the recommended benchmark vector through the collaborative filtering recommendation mechanism. Combine the task preference vector and the recommended benchmark vector to portray the user profile to obtain the user preference feature vector; Generate a recommendation list: Calculate the similarity between the topic feature vector and the user preference feature vector through the cosine similarity function, use the similarity as the recommendation factor, and generate a recommendation list according to the recommendation factor and the recommendation threshold.

[0037] Specifically, the cosine similarity is defined as the following expression:

[0038] where, is the similarity, A and B respectively represent the user preference feature vector and the topic feature vector, and respectively represent the i-th user preference feature and topic feature in the user preference feature vector and the topic feature vector, x is the total number of user preference features in the user preference feature vector, and also the total number of topic features in the topic feature vector.

[0039] In the above steps, the satisfaction degree of the user with the topic recommendation list will be fed back to the user feature learning process in the form of scores. By continuously accumulating the score data of the user's feedback on the topics, the user preference features can be updated in a timely manner, so as to improve the accuracy of topic recommendation for the user.

[0040] The task preference vector can describe the preference degree of the user for each topic message, while the recommended benchmark vector can describe the preference degree of a similar user group for each topic message. By measuring the deviation between each user's task preference vector and the recommended benchmark vector, and using this deviation together with the task preference vector as the basis for the recommendation factor, the purpose of moderately correcting the individual data recommendation result by using the group data access behavior can be achieved, making the recommendation factor more accurate and the topic message recommended for the user more effective.

[0041] Optionally, the features of the topics in the topic message queue are extracted through the TF-IDF keyword extraction technology, which is the topic feature vector, including: The basic idea of the TF-IDF method is as follows: on the one hand, the more times the keyword k appears in the document D, the more important the keyword k is to the document D, and the more capable it is to represent the semantics of the document D; on the other hand, the more times the keyword k appears in different documents, the less contribution k makes to differentiating documents. Considering the above two aspects comprehensively, the feature weight setting method of TF-IDF is proposed as follows: The topics in the topic message queue are a set containing N topic message data, that is, { }, among which the message data containing the keyword is ones, represents the keyword appears in the message data times, then in the message data the word frequency is defined as the number of times appears in the total number of times all z keywords appear in the message data , that is:

[0042] Among them, z represents the number of keywords that appear in the message data , that is, there are keywords { } in the message data , represents the keyword appears in the message data times, represents the sum of the number of times all keywords appear in the message data; The inverse frequency that appears in the topic is defined as:

[0043] where represents the total number of messages in the topic, represents the number of files containing the keyword ; The TF-IDF parameter is defined as:

[0044] Adopt a k-dimensional vector to represent the characteristics of the topic in the topic message queue.

[0045] Optionally, in feature learning: Obtain the user's historical access records and historical subscription records in recent tasks, and combine the user's feedback score results on intelligent recommendation topics; extract the user's interest topic set from the historical records through the LDA algorithm, and construct it into a task preference vector; In order to more accurately depict the user profile, obtain a recommendation benchmark vector through a collaborative filtering recommendation mechanism; Use the task preference vector as the main part of calculating the user preference feature vector, and use the deviation between each user's task preference vector and the recommendation benchmark vector as the secondary part of calculating the user preference feature vector to obtain the user preference feature vector.

[0046] In the above embodiment, in the feature learning process, when the intelligent recommendation mechanism sets the recommendation factor for the user and the topic message queue, a user profile construction method based on the Linear Discriminant Analysis (LDA) algorithm is adopted, that is, through a given training example set, project the training examples onto a straight line, so that the projection points of the same type of examples are as close as possible, and the projection points of different types of examples are as far away as possible. When classifying new samples, project them onto the same straight line, and determine the category of the new samples according to the position of the projection points. Combining the characteristics of the single user role and limited user subscription data volume in the airborne distributed environment, a content-based recommendation algorithm is adopted. Content recommendation is the most basic algorithm of the recommendation system, but until now it is still the most effective and direct algorithm, and is applicable to various fields. The intelligent recommendation mechanism in this application will collect the user's data access history records in recent tasks, as well as the user's historical subscription records, and combine the user feedback score as the training example set, extract the user's interest topic set from the access history through the LDA algorithm, and construct a task preference vector for it.

[0047] Optionally, the recommended reference vector obtained through the collaborative filtering recommendation mechanism includes: Calculate similarity: Based on the task preference vector of each user, calculate the similarity between the target user and other users through cosine similarity; Identify similar user groups: Determine a similarity threshold or select the top N approximate users as the similar user group; Determine the recommended topic: Collect the set of interest topics of the similar user group of the target user, including the subscribed and visited topics, to form the recommended reference vector.

[0048] Optionally, the update method of the recommendation factor includes: Input: User preference feature P, feature learning algorithm A, all topic information S; 1) : Extract the topic feature vector from the topic information through the TF-IDF keyword extraction method; is the topic feature vector, is the extraction function; 2) Compare the user preference feature P with the topic feature vector IR_S, use the VSM model, calculate the similarity between P and IR_S, sort the similarities, that is, sort the recommendation factors, and obtain the recommendation list R1; 3) The user scores the recommended information in the recommendation list R1, including explicit feedback and implicit feedback, and the feedback information is t; the explicit feedback is the scoring information; the implicit feedback is the historical access information; 4) Update the user preference feature , is the update function, t is the time, to improve the recommendation accuracy; 5) Return to step 2) and continue to generate the recommendation list R1; Output: The recommendation list R1 of the user, the user preference feature P.

[0049] Optionally, verify the effectiveness of the recommendation factor through relevant indicators; the relevant indicators include coverage, discoverability, and scoring.

[0050] Coverage: Represents the ability of the intelligent recommendation mechanism to discover message topics. The coverage is defined as:

[0051] Among them, N represents the total number of user message topics, represents the topic list recommended by the intelligent recommendation mechanism to the user, represents the number of topics included in the recommended topic list, is the topic representing the recommended topic list to the user; Discoverability: It refers to the ability of the intelligent recommendation mechanism to recommend to users the message data that the users have not subscribed to but is suitable for the users' topics. The discoverability is defined by the following expression:

[0052] where n is the number of historical recommendations counted, is the number of topics that were not selected before but were selected in the historical selection of a certain recommendation, is the number of topics that were selected in this recommendation and were also selected before; Score: It refers to the degree of satisfaction of users with the topic list recommended by the intelligent recommendation mechanism. The score is evaluated using precision and recall. The definition of precision is:

[0053] The definition of recall is:

[0054] where, is the precision, is the recall, represents the list of topics that the user truly needs, is all the topics in the topic message queue.

[0055] For the convenience of understanding, this application gives a more specific embodiment based on the above embodiments: Refer to Figure 1 , in an optimal embodiment described below, take the publishing and subscribing process of User 1 and the intelligent recommendation mechanism pushing topics to User 4 as an example. In the data publishing process, Users 1, 2, 3, and 4 divide the collected sensor data according to different topics, namely Topic 1, Topic 2, Topic 3, and Topic 4, and push the classified data to the corresponding topic message queues through the publish / subscribe management function. In the data subscribing process, User 1 subscribes to Topic 4. Therefore, User 1 can actively obtain the data of Topic 4 from the topic message queue through the publish / subscribe management function. In the intelligent recommendation process, the intelligent recommendation mechanism extracts the user preference feature vectors based on the user's historical subscription records, historical access records, and user feedback scores, obtains the topic feature vectors of all topics from the topic message queue, calculates the recommendation factor through the intelligent recommendation mechanism, and then determines whether the recommendation factor is greater than the recommendation threshold set by the user. For example, if the recommendation factor is greater than the threshold set by User 4, the data topic push function will push the result to User 4. If User 4 determines that the data is the information he / she is interested in, he / she can initiate an acquisition request to obtain the topic data from the topic message queue.

[0056] Refer toFigure 2 , which is the schematic diagram of the intelligent push mechanism of the present invention, and its detailed process is as follows: The oval box represents two entities, including the topic message queue and the user. Among them, the topic message queue is the data producer; the user is the data consumer.

[0057] The rounded rectangle box represents important intermediate results that need to be stored, including topic features, user preference features, topic recommendation lists, historical subscription records, historical access records, and ratings of recommended data. Among them, topic features represent the characteristics of data topics; user preference features represent the preferences of users for subscribed topics; topic recommendation lists represent the final recommended results generated by the intelligent recommendation mechanism for recommendation to users; historical subscription records represent the subscriptions of users to data topics through the publish-subscribe module; historical access records and ratings of recommended data represent the feedback of users on the recommended topic list.

[0058] The rectangular box represents three steps of the recommendation algorithm, including topic representation, feature learning, and generating a recommendation list. Among them, topic representation refers to the process of extracting each topic feature from the topic message queue; feature learning refers to the process of extracting user preference features that can represent user preferences from user subscription records, user access records, and user ratings of recommended data.

[0059] The small file-shaped box represents the algorithms or processes in the intelligent recommendation mechanism process, including TF-IDF parsing, Item structured filtering, data subscription management, linear discriminant analysis, task preference vector and recommendation benchmark vector, user feedback on the recommended result, cosine function, recommended result, and recommendation. Among them, TF-IDF parsing and Item structured filtering represent the most important processes of extracting topic features by performing the topic representation step from the topic message queue; data subscription management represents the process of generating historical subscription records according to the user's subscription operation; linear discriminant analysis represents the process of extracting the user task preference vector and recommendation benchmark vector by performing the feature learning process from historical subscription records, historical access records, and user ratings of recommended data, and constructing user preference features; the cosine function method represents the process of the intelligent recommendation mechanism generating a recommendation list based on topic features and user preference features; recommended result and recommendation represent the process of pushing the topic recommendation list to the user.

[0060] Through the user's feedback on the topics in the recommendation list, the user preference features can be updated in a timely manner. After the recommendation factors are fed back by the user, they will be continuously updated, which can depict a more accurate portrait of the user and achieve better recommendation effects.

[0061] Referring to Table 1, in one preferred embodiment described below, the intelligent recommendation mechanism is designed to provide users with an actively pushed data service. At the same time, the data transmission is jointly determined by the intelligent recommendation mechanism and the user, reducing unnecessary data transmission and occupying less bandwidth. Assume that the length of a single piece of data is L, the number of data to be transmitted is M, the length of the message notification is B, the recommendation rate of the subscription management module is α, and the probability that the user obtains data after receiving the message notification is p. Then, the transmission bandwidth consumption for the intelligent recommendation process of this application is . When the data length is much larger than the notification message length, i.e., L >> B, .

[0062] Table 1 Analysis and Comparison of Transmitted Data Volume

[0063] Compared with the active data query scheme, users can only obtain the required data through keyword retrieval or range retrieval, cannot filter data through the recommendation system, and cannot decide whether to obtain data by themselves. Therefore, its bandwidth only depends on the query relevance γ and the number of queries n, that is .

[0064] Generally, since the recommendation rate of the intelligent recommendation mechanism can be continuously updated according to user feedback scores and user historical subscription records, and the query relevance γ is determined by the user's own filtering conditions, and its filtering range is more ambiguous, so the γ generated by the query conditions is greater than the α adopted by the intelligent recommendation. Through comparative analysis, it can be seen that implementing the intelligent recommendation process of airborne data with this application can effectively reduce the occupation of airborne bandwidth resources.

[0065] Referring to Table 1, in one preferred embodiment described below, for the model operation settings and process display, an existing offline dataset is used to test and evaluate the recommendation algorithm. During the test, 100 users are recommended topics, the length of the recommendation list is 10, the total number of topics is 10, and the total number of corresponding data files is 240. Through the parameter settings of the algorithm, the calculated coverage rate and popularity are 45% and 5% respectively. For the calculation results of the accuracy of the recommendation results, there are 245 correctly recommended data files and 36 un-recommended data files. The accuracy rate reaches 24.5%, and the recall rate is 87.2%. The recommendation effect is relatively accurate, and the accuracy rate and recall rate can basically meet the expectations. Through continuous feedback information from users, the accuracy of the recommendation algorithm can be further improved.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific implementation manners of the present application, and any modification or equivalent replacement that does not depart from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. An intelligent distributed interaction method for an airborne environment, characterized in that Including: The data publisher classifies the relevant data it obtains by theme and uploads the classified data to the theme message queue of the server for users to obtain; the relevant data includes message data and environmental data; the user is an airborne platform; The user obtains data from the theme message queue according to its own needs.

2. The method according to claim 1, wherein The user obtains data from the theme message queue of the server according to its own needs, including: For the themes that the user has subscribed to before, the user directly obtains data from the theme message queue through the publish / subscribe management function of the server without the need for recommendation by the intelligent recommendation mechanism of the server; For the themes that the user has not subscribed to before, the data theme push function of the server establishes an association between each user and each theme. Based on the intelligent recommendation mechanism, the degree of interest of each user in each theme is measured. According to the user's historical subscription records, historical access records, and user feedback scores, the user preference feature vector is extracted, and the similarity between the user preference feature vector of each user and the theme feature vector of each data theme is calculated, and this similarity is used as the recommendation factor; a recommendation function other than the data publish / subscribe service is provided for the user through the recommendation factor.

3. The method according to claim 2, wherein The providing of a recommendation function other than the data publish / subscribe service for the user through the recommendation factor includes: When the recommendation factor is greater than or equal to the recommendation threshold, it indicates that the recommended theme is valuable to the user. The recommended theme is added to the recommendation list, and a recommendation message is sent to the user; when the recommendation factor is less than the recommendation threshold, the intelligent recommendation mechanism ignores the recommended theme and waits to re-judge whether to recommend the theme when the user preference changes.

4. The method according to claim 3, characterized in that, After the user receives the recommended theme, if the recommended theme is a theme that the user is interested in, a request is initiated to the data theme push function of the server. The data theme push function of the server actively obtains the corresponding message data from the theme message queue. The access information of the user to the data theme push function of the server, that is, the historical access record, will be fed back to the user preference to affect the portrayal of the user profile; when the recommended theme is not a theme that the user is interested in, the user's feedback will also affect the next portrayal of the user profile.

5. The method according to claim 3, wherein The intelligent recommendation mechanism includes: Theme representation: Extract the features of the themes in the theme message queue through the TF-IDF keyword extraction technology, which is the theme feature vector, and use this theme feature vector to represent the recommended theme; Feature learning: According to the user's historical access and subscription theme records, as well as the feedback scores on the recommended themes, obtain the task preference vector through the LDA algorithm, obtain the recommendation benchmark vector through the collaborative filtering recommendation mechanism, and combine the task preference vector and the recommendation benchmark vector to portray the user profile to obtain the user preference feature vector; Generate a recommendation list: Calculate the similarity between the theme feature vector and the user preference feature vector through the cosine similarity function, use the similarity as the recommendation factor, and generate a recommendation list according to the recommendation factor and the recommendation threshold.

6. The method according to claim 5, characterized in that, The extracting of the features of the themes in the theme message queue through the TF-IDF keyword extraction technology, which is the theme feature vector, includes: The topic in the topic message queue is a set containing N topic message data, that is, { }, which contains the keyword The number of message data is pieces, Indicates the keyword The number of times it appears in the message data Then In the message data The word frequency Is defined as the total number of occurrences of all z keywords in the message data The number of times Appears, that is: where z represents the number of keywords that appear in the message data i.e., the number of keywords that exist in the message data { }, represents the keyword appears in the message data the number of times; represents the sum of the number of times all keywords appear in the message data; The inverse frequency that appears in the subject is defined as: Among them, represents the total number of messages in the subject, represents the number of files containing the keyword ; TF-IDF parameter is defined as: Use a k-dimensional vector to represent the characteristics of the topic in the topic message queue.

7. The method according to claim 5, characterized in that, In the said feature learning: Obtain the user's historical access records and historical subscription records in recent tasks, and combine the user's feedback score results on the intelligent recommendation topics; extract the user's interest topic set from the historical records through the LDA algorithm, and construct it into a task preference vector; The recommended benchmark vector obtained through the collaborative filtering recommendation mechanism; Use the task preference vector as the main part of calculating the user preference feature vector, and use the deviation between each user's task preference vector and the recommended benchmark vector as the secondary part of calculating the user preference feature vector to obtain the user preference feature vector.

8. The method according to claim 5 or 7, characterized in that The recommended benchmark vector obtained through the collaborative filtering recommendation mechanism includes: Calculate similarity: Based on each user's task preference vector, calculate the similarity between the target user and other users through cosine similarity; Identify similar user groups: Determine a similarity threshold or select the top N approximate users as similar user groups; Determine recommended topics: Collect the interest topic sets of the target user's similar user groups, including subscribed and accessed topics, to form a recommended benchmark vector.

9. The method according to any one of claims 2-4, characterized in that, The update method of the recommended factor includes: Input: User preference feature P, feature learning algorithm A, all topic information S; 1) : Extract the topic feature vector from the topic information through the TF-IDF keyword extraction method; is the topic feature vector; is the extraction function; 2) Compare the user preference feature P with the topic feature vector IR_S, use the VSM model, calculate the similarity between P and IR_S, sort the similarities, that is, sort the recommended factors, to obtain the recommended list R1; 3) The user scores the recommended information in the recommended list R1, there are explicit feedback and implicit feedback, and the feedback information is t; the explicit feedback is the scoring information; the implicit feedback is the historical access information; 4) Update user preference features , is an update function, t is time, to improve recommendation accuracy; 5) Return to step 2) and continue to generate the recommended list R1; Output: The user's recommended list R1, user preference feature P.

10. The method according to any one of claims 2-4, characterized in that, Verify the effectiveness of the recommended factor through relevant metrics; the relevant metrics include coverage, discoverability, and scoring; Coverage: Represents the ability of the intelligent recommendation mechanism to explore message topics. The coverage is defined as: where N represents the total number of user message topics, represents the list of topics recommended to the user by the intelligent recommendation mechanism, represents the number of topics included in the recommended topic list, represents the topic of the list of topics recommended to the user; Discoverability: Represents the ability of the intelligent recommendation mechanism to recommend to the user the message data that the user has not subscribed to but is suitable for the user's topic. The discoverability is defined by the following expression: where n is the number of historical recommendations counted, is the number of topics that were not selected in previous historical selections for a certain recommendation but were selected this time, is the number of topics that were selected in this recommendation and were also selected before; Scoring: Represents the user's satisfaction with the topic list recommended by the intelligent recommendation mechanism; the scoring is evaluated using precision and recall. The precision is defined as: The recall is defined as: Among them, is the accuracy rate, is the recall rate, R(u) represents the list of topics recommended to the user by the intelligent recommendation mechanism, represents the list of topics that the user truly needs, is all the topics in the topic message queue.

Images