Real-time place recommendation method and system based on combined empowerment and user feedback

Abstract

The invention discloses a site real-time recommendation method based on combined empowerment and user feedback, which comprises the following steps: carrying out feature extraction on site related data of three modalities of eye movement data, operation records and text data within a period of time, and obtaining a site attention ranking list under each modality; initializing a subjective weight; calculating an objective weight; a difference coefficient method is adopted to integrate the subjective weight and the objective weight, and an output place recommendation list is calculated; feedback information of a user to the site recommendation list is obtained, the normalized loss accumulation gain of the feedback information and the site attention ranking list in different modes is calculated, and the relative importance degree and subjective weight of three kinds of mode information are fed back and updated. According to the method, the data from the multivariate acquisition system can be identified, judged, comprehensively processed and fused to obtain more comprehensive, accurate and reliable information than any single data source, so that the inherent defects of the single data source are effectively overcome, and high-coverage and high-precision information is generated.

Description

Technical Field

[0001] The present invention particularly relates to the field of personalized recommendation systems, and in particular, provides a method and system for real-time location recommendation based on multimodal data fusion and dynamic weight adjustment when a user interacts with a geographic information system. The system provides users with accurate location recommendation services by integrating user eye movement data, operation behavior and text input, and is particularly suitable for map services and location-based services that require fast and dynamic response to user needs. Background Art

[0002] Existing recommendation systems often face the problem of large amounts of data, diverse types, and complexity when processing data from multiple acquisition systems. Although data fusion technology has solved the problem of information integration to a certain extent, existing methods often lack the ability to dynamically adjust data weights and make personalized recommendations. This results in the recommendation results being inaccurate and unable to meet users' high requirements for real-time and personalization. The present invention significantly improves the relevance and accuracy of location recommendations by introducing a dynamic weight adjustment mechanism based on user behavior and an adaptive learning algorithm combined with user feedback, thereby solving the specific difficulties of slow response and low personalization of recommendation systems in the prior art. Summary of the invention

[0003] The purpose of the present invention is to disclose a real-time location recommendation method and system based on combined weighting and user feedback, which identifies, judges, comprehensively processes and integrates data from multiple acquisition systems to obtain more comprehensive, accurate and reliable information than any single data source, thereby effectively making up for the inherent deficiencies of a single data source, generating high-coverage, high-precision information, and effectively meeting the needs of command and dispatch and public services.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0005] A method for real-time location recommendation based on combined weighting and user feedback, the method comprising the following steps:

[0006] S1: Extract features from the user's eye movement data, mouse and keyboard operation records, and text data over a period of time. The eye movement data is used to extract the attention of the location, the operation records are used to extract the click frequency of the location, and the text data is used to extract the keyword frequency of the location, so as to generate an initial location attention ranking list of the three modes;

[0007] S2: Based on expert experience and domain knowledge, the G1 algorithm is used to determine the relative importance of eye movement data, operation records and text data, and initialize the subjective weight of each modal information;

[0008] S3: Use the CRITIC algorithm to comprehensively analyze the location attention ranking list of eye movement data, operation records and text data, and calculate the objective weights of the three modal data, where the objective weights reflect the internal consistency of each modal data and its contribution to the overall decision;

[0009] S4: dynamically merge the subjective weight in step S2 and the objective weight in step S3 using the difference coefficient method to generate a combined weight, wherein the calculation of the combined weight combines the relative importance and timeliness of different modal information in different scenarios, thereby calculating and outputting a comprehensive location recommendation list;

[0010] S5: Obtain user feedback on the place recommendation list, use the Normalized Discounted Cumulative Gain (NDCG) to evaluate the ranking accuracy of each modal information, and update the relative importance and subjective weight of the three modal information through a feedback loop to ensure that the recommendation system responds promptly to changes in user preferences;

[0011] S6: Repeat steps S3 to S5 at preset time intervals, dynamically update the subjective weights and the location recommendation list while continuously obtaining user feedback, and gradually enhance the adaptive capability of the recommendation system in different application scenarios to improve recommendation accuracy.

[0012] Step S1 further comprises:

[0013] Extract features from eye movement data: Use the DBSCAN clustering algorithm to cluster the collected gaze points to obtain the user's area of ​​interest; obtain the place name of each cluster center by calling the background map database, count the number of gaze points in each cluster to determine the user's attention to the place corresponding to the area of ​​interest; output the place attention ranking list in the eye movement data mode from high to low according to the place attention;

[0014] Extract features from operation records: obtain a ranking list of locations in the operation record mode by recording the names and times of mouse clicks and keyboard searches through background monitoring;

[0015] Feature extraction of text data: Jieba word segmentation algorithm and part-of-speech tagging are used to count the place names and frequencies in the text to obtain a ranked list of place attention under the text data mode.

[0016] Step S2 further comprises:

[0017] S21, sorting each modal information from high to low according to the degree of importance based on expert experience knowledge, and obtaining the order relationship of each modal information;

[0018] S22, for any two adjacent modal information in the order relationship, the decision maker assigns relative importance where λ j is the weight of the jth modal information, r j The larger the modal information compared to More important;

[0019] S23, according to the relative importance of each modal information determined by the decision maker, the subjective weight of each modal information is obtained by the following formula:

[0020]

[0021] λ j-1 =λ j r j

[0022] Where m is the total amount of modal information, and

[0023] Step S3 further comprises:

[0024] S31, constructing a data matrix x according to the ranking list of the location attention under the three model information calculated in step S1, where n is the total number of modal types, m is the total number of locations, and x ij Indicates the attention level of the i-th location under the j-th modal information:

[0025]

[0026] S32, using the min-max normalization method to perform dimensionless processing on the location attention data of each modal information to obtain a normalized data matrix X;

[0027] S33, calculate the standard deviation of each modal data to represent the difference fluctuation of each modal information, and the standard deviation calculation formula is as follows:

[0028]

[0029] Where S j is the standard deviation of the j-th modal data;

[0030] S34, calculating the correlation coefficient between each modal data to indicate the degree of conflict between each modal information, the correlation coefficient calculation formula is as follows:

[0031]

[0032] where r jj' is the correlation coefficient between the jth modal data and the j′th modal data, σ j ,σ j' are the standard deviations of the jth modal data and the j′th modal data, μ j ,μ j'are the means of the j-th modal data and the j′-th modal data respectively.

[0033] S35, then the conflict calculation formula of j modal information is as follows:

[0034]

[0035] S36, calculating the amount of information of each modal information:

[0036] C j =S j *R j

[0037] C j The larger the value, the more weight should be assigned to the jth modal information.

[0038] S36, calculate the objective weight of each modal information:

[0039]

[0040] Step S4 further comprises:

[0041] S41, setting subjective weight λ j and objective weight ω j The coefficient of and ξ, the combined weights are as follows:

[0042]

[0043] S42, use the difference coefficient method to solve the coefficient and ξ,

[0044]

[0045] Where P j (i=1,2,...,m), which is the vector of subjective weights sorted in ascending order, and m is the total amount of modal information.

[0046] Step S5 further comprises:

[0047] Calculate the NDCG value of the feedback information and the sorted list of each modal information:

[0048] ndcg1=NDCG(f,m)

[0049] ndcg2=NDCG(f,n)

[0050] ndcg3=NDCG(f,e)

[0051] Where f is the feedback information, m, n and e are the ranked lists of place attention in the three modes of eye movement data, operation records and text data respectively, and ndcg1, ndcg2 and ndcg3 are the NDCG values ​​calculated for the feedback information and the ranked lists of each mode respectively.

[0052] Sort the NDCG values ​​from large to small to determine the order relationship of the three modal information:

[0053] sortndcg=sort(ndcg1,ndcg2,ndcg3)

[0054] Among them, sort(·) sorts the NDCG values ​​from large to small, and the output sortndcg includes the NDCG values ​​represented by each modal information after sorting, which are sortndcg(0), sortndcg(1) and sortndcg(2) respectively;

[0055] Calculate the relative importance of the three modal information:

[0056]

[0057] The relative importance of the three modal information is calculated based on the order relationship and NDCG value, and the subjective weight can be updated accordingly.

[0058] In a second aspect, a real-time location recommendation system based on combined weighting and user feedback is provided, the real-time location recommendation system comprising:

[0059] Data feature extraction unit: used to extract features from the location-related data in the three modes of eye movement data, operation records and text data within a period of time, and obtain a ranking list of location attention in each mode;

[0060] Subjective weight generation unit: uses the G1 algorithm to determine the relative importance of the three modal information to initialize the subjective weights of the three modal information;

[0061] Objective weight generation unit: Use the CRITIC algorithm to comprehensively analyze the ranking list of location attention under each mode and calculate the objective weights of the three modal information;

[0062] Comprehensive weight generation unit: uses the difference coefficient method to combine the subjective weight and the objective weight to obtain the combined weight, and calculates and outputs the location recommendation list based on the combined weight;

[0063] User feedback generation unit: obtains user feedback information on the place recommendation list, calculates the normalized discounted cumulative gain NDCG between it and the place attention ranking list under different modes, and updates the relative importance and subjective weight of the three modal information.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The real-time location recommendation method and system based on combined weighting and user feedback of the present invention identifies, judges, comprehensively processes and integrates data from multiple acquisition systems to obtain more comprehensive, accurate and reliable information than any single data source, thereby effectively making up for the inherent deficiencies of a single data source, generating high-coverage, high-precision information, and effectively meeting the needs of command, dispatch and public services. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flow chart of a method for real-time location recommendation based on combined weighting and user feedback of the present invention;

[0067] Figure 2 Schematic diagram of weighting the combination;

[0068] Figure 3 Schematic diagram of feedback mechanism;

[0069] Figure 4 Schematic diagram of the comparison of NDCG value curves with and without feedback mechanism. DETAILED DESCRIPTION

[0070] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0071] The present invention discloses a method for real-time location recommendation based on combined weighting and user feedback, and the method comprises the following steps:

[0072] S1: Extract features from the user's eye movement data, mouse and keyboard operation records, and text data over a period of time. The eye movement data is used to extract the attention of the location, the operation records are used to extract the click frequency of the location, and the text data is used to extract the keyword frequency of the location, so as to generate an initial location attention ranking list of the three modes;

[0073] S2: Based on expert experience and domain knowledge, the G1 algorithm is used to determine the relative importance of eye movement data, operation records and text data, and initialize the subjective weight of each modal information;

[0074] S3: Use the CRITIC algorithm to comprehensively analyze the location attention ranking list of eye movement data, operation records and text data, and calculate the objective weights of the three modal data, where the objective weights reflect the internal consistency of each modal data and its contribution to the overall decision;

[0075] S4: dynamically merge the subjective weight in step S2 and the objective weight in step S3 using the difference coefficient method to generate a combined weight, wherein the calculation of the combined weight combines the relative importance and timeliness of different modal information in different scenarios, thereby calculating and outputting a comprehensive location recommendation list;

[0076] S5: Obtain user feedback on the place recommendation list, use the Normalized Discounted Cumulative Gain (NDCG) to evaluate the ranking accuracy of each modal information, and update the relative importance and subjective weight of the three modal information through a feedback loop to ensure that the recommendation system responds promptly to changes in user preferences;

[0077] S6: Repeat steps S3 to S5 at preset time intervals, dynamically update the subjective weights and the location recommendation list while continuously obtaining user feedback, and gradually enhance the adaptive capability of the recommendation system in different application scenarios to improve recommendation accuracy.

[0078] See also Figure 1 The real-time location recommendation method of the present invention specifically includes the following steps:

[0079] Step 1: feature extraction of eye movement data includes: first, clustering the gaze points collected over a period of time using the DBSCAN clustering algorithm to obtain the user's area of ​​interest (cluster). Then, the name of the place at the center of each cluster is obtained by calling the background map database. Next, the number of gaze points in each cluster is counted to determine the user's attention to the place. Finally, the place attention ranking list under this modality is output from high to low according to the attention of the place.

[0080] The feature extraction operation of the operation record includes: obtaining a ranking list of the place attention in the mode by recording the place names and times of mouse clicks and keyboard searches within a period of time through background monitoring;

[0081] The feature extraction operation of text data includes: using Jieba word segmentation algorithm and part-of-speech tagging to count the place names and frequencies in the text to obtain a ranked list of place attention under this mode.

[0082] Step 2: Use the G1 algorithm to subjectively determine the relative importance of the three modal information to initialize the subjective weight allocation, as shown in the reference Figure 2 As shown, the following steps are included:

[0083] 1) Determine the order relationship of each modal information:

[0084] According to the expert experience knowledge, the modal information is sorted from high to low in terms of importance to obtain the order relationship Examples are shown in Table 1.

[0085] Table 1 Determine the order relationship of each modal information

[0086]

[0087] 2) Determine the relative importance of adjacent modal information in the order relationship:

[0088] For any two adjacent modal information and The relative importance assigned by decision makers can be expressed as where j = 1, 2, 3, λ j is the weight of the jth modal information, r j The larger the modal information compared to More important. j Refer to Table 2 for the value of:

[0089] Table 2 Relative importance r j Value table

[0090]

[0091] Based on the above table and the order relationship determined in 1), it is subjectively believed that operation records are slightly more important than text data, and text data is slightly more important than eye movement data, as shown in Table 3.

[0092] Table 3 Relative importance of modal information

[0093]

[0094] 3) Calculate the subjective weight of each modal information:

[0095] According to the decision maker's j The subjective weight of each modal information can be obtained by the following formula.

[0096]

[0097] λ j-1 =λ j r j

[0098] where j = 1, 2, 3, and According to 2) the relative importance r is calculated j The weights of the three modal information are calculated as shown in the following table.

[0099] Table 4 Weight of each modal information

[0100] Modal Information Operation log Text Data Eye tracking data Weight 0.39 0.33 0.28

[0101] In step 3, the CRITIC algorithm is used to calculate the objective weights of the three modal information, as shown in reference Figure 2 As shown, the following steps are included:

[0102] 1) Dimensionless processing:

[0103] According to the ranking list of the location attention under the three model information calculated in step 1, a data matrix x is constructed, where n is the total number of modal types, m is the total number of locations, and x is the total number of locations. ij Indicates the attention level of the i-th location under the j-th modal information:

[0104]

[0105] In order to eliminate the influence of different dimensions on the data, it is necessary to process the data of each modal information dimensionlessly. This patent uses the min-max normalization method.

[0106]

[0107] 2) Calculate the fluctuation of each modal information:

[0108] In the CRITIC weighting method, the standard deviation is used to represent the difference fluctuation of each modal information. The larger the standard deviation, the greater the difference of the modal information, and the more information it can reflect. Therefore, more weight should be assigned to the modal information. The standard deviation calculation formula is as follows:

[0109]

[0110] Where S j is the standard deviation of the j-th modal information.

[0111] 3) Calculate the conflict degree of each modal information:

[0112] The degree of conflict uses the correlation coefficient to represent the correlation between each modal information. The higher the correlation, the more identical information there is and the more repetitive the data is. This weakens the data value of the modal information to a certain extent, and the weight assigned to the modal information should be reduced.

[0113]

[0114] where r jj' is the correlation coefficient between the jth modal information and the j′th modal information, σ j ,σ j' are the standard deviations of the jth modal information and the j′th modal information, μ j ,μ j' are the means of the j-th modal information and the j′-th modal information respectively.

[0115] The conflict calculation formula for the jth modal information is as follows:

[0116]

[0117] 4) Calculate the amount of information of each modal information:

[0118] C j =S j *R j

[0119] The larger the value is, the greater the role of the j-th modal information in the overall comprehensive decision-making, and more weight should be assigned to it.

[0120] 5) Calculate the objective weight of each modal information:

[0121]

[0122] In step 4, set the subjective weight λ j and objective weight ω j The coefficient of and ξ, the combined weights are as follows:

[0123]

[0124] Then the coefficients are solved by using the difference coefficient method and ξ,

[0125]

[0126] Where P j (i=1,2,...,m), which is the vector of subjective weights sorted in ascending order, and m is the total amount of modal information.

[0127] In step 5, as referenced Figure 3 As shown, the NDCG value is calculated for the user's feedback information and the sorted lists of the three modal information respectively. The larger NDCG value indicates that the sorted list of the modal information is more in line with the user's intention. When updating the weight, the weight of the acquisition method can be increased and the weight of other methods can be reduced. Therefore, the order relationship of each modal information can be determined based on the comparison of the NDCG values ​​of the three, and then the ratio of the larger NDCG value to the smaller NDCG value is used to represent the relative importance of adjacent modal information in the order relationship. The specific process is as follows:

[0128] 1) Calculate the NDCG value of the feedback information and the sorted list of each modal information:

[0129] ndcg1=NDCG(f,m)

[0130] ndcg2=NDCG(f,n)

[0131] ndcg3=NDCG(f,e)

[0132] Where f is the feedback information, m, n and e are the ranking lists of the location attention in the three modes of eye movement data, operation records and text data, respectively, and ndcg1, ndcg2 and ndcg3 are the NDCG values ​​calculated from the feedback information and the ranking lists of each mode. 2) Sort the NDCG values ​​from large to small to determine the order relationship of the three modal information:

[0133] sortndcg=sort(ndcg1,ndcg2,ndcg3)

[0134] Among them, sort(·) sorts the NDCG values ​​from large to small, and the output sortndcg includes the NDCG values ​​represented by each modal information after sorting, which are sortndcg(0), sortndcg(1) and sortndcg(2) respectively;

[0135] 3) Calculate the relative importance of the three modal information:

[0136]

[0137] The relative importance of the three modal information is calculated based on the order relationship and NDCG value, and the subjective weight can be updated accordingly.

[0138] In step six, repeat steps three to five and update the comprehensive ranking list of subjective weights and location attention at regular intervals.

[0139] According to another aspect of the present invention, there is also provided a method and system for real-time location recommendation based on combined weighting and user feedback, comprising:

[0140] Data feature extraction unit: used to extract features from the location-related data in the three modes of eye movement data, operation records and text data within a period of time, and obtain a ranking list of location attention in each mode;

[0141] Subjective weight generation unit: uses the G1 algorithm to determine the relative importance of the three modal information to initialize the subjective weights of the three modal information;

[0142] Objective weight generation unit: Use the CRITIC algorithm to comprehensively analyze the ranking list of location attention under each mode and calculate the objective weights of the three modal information;

[0143] Comprehensive weight generation unit: uses the difference coefficient method to combine the subjective weight and the objective weight to obtain the combined weight, and calculates and outputs the location recommendation list based on the combined weight;

[0144] User feedback generation unit: obtains user feedback information on the place recommendation list, calculates the normalized discounted cumulative gain NDCG between it and the place attention ranking list under different modes, and updates the relative importance and subjective weight of the three modal information.

[0145] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0146] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0149] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0150] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A real-time location recommendation method based on combined weighting and user feedback, characterized in that: The real-time location recommendation method comprises the following steps: S1: Extract features from the user's eye movement data, mouse and keyboard operation records, and text data over a period of time. The eye movement data is used to extract the attention of the location, the operation records are used to extract the click frequency of the location, and the text data is used to extract the keyword frequency of the location, so as to generate an initial location attention ranking list of the three modes; S2: Based on expert experience and domain knowledge, the G1 algorithm is used to determine the relative importance of eye movement data, operation records and text data, and initialize the subjective weight of each modality information; S3: Use the CRITIC algorithm to comprehensively analyze the location attention ranking list of eye movement data, operation records and text data, and calculate the objective weights of the three modal data, where the objective weights reflect the internal consistency of each modal data and its contribution to the overall decision; S4: dynamically merge the subjective weight in step S2 and the objective weight in step S3 using the difference coefficient method to generate a combined weight, wherein the calculation of the combined weight combines the relative importance and timeliness of different modal information in different scenarios, thereby calculating and outputting a comprehensive location recommendation list; S5: Obtain user feedback on the place recommendation list, use the normalized discounted cumulative gain (NDCG) to evaluate the ranking accuracy of each modal information, and update the relative importance and subjective weight of the three modal information through a feedback loop to ensure that the recommendation system responds promptly to changes in user preferences; S6: Repeat steps S3 to S5 at preset time intervals, dynamically update the subjective weights and the location recommendation list while continuously obtaining user feedback, and gradually enhance the adaptive capability of the recommendation system in different application scenarios to improve recommendation accuracy.

2. The method for real-time location recommendation based on combined weighting and user feedback according to claim 1, characterized in that: Step S1 further comprises: Extract features from eye movement data: Use the DBSCAN clustering algorithm to cluster the collected gaze points to obtain the user's area of ​​interest; obtain the place name of each cluster center by calling the background map database, count the number of gaze points in each cluster to determine the user's attention to the place corresponding to the area of ​​interest; output the place attention ranking list in the eye movement data mode from high to low according to the attention of the place; Extract features from operation records: obtain a ranked list of locations in the operation record mode by recording the location names and frequencies of mouse clicks and keyboard searches through background monitoring; Extract features from text data: Use word segmentation algorithms and part-of-speech tagging to count the keyword frequencies of places in the text and obtain a ranked list of place attention under the text data modality.

3. The method for real-time location recommendation based on combined weighting and user feedback according to claim 1, characterized in that: Step S2 further comprises: S21, sorting each modal information from high to low according to the degree of importance based on expert experience knowledge, and obtaining the order relationship of each modal information; S22, for any two adjacent modal information in the order relationship, the decision maker assigns relative importance where λ j is the weight of the jth modal information, r j The larger the modal information compared to More important; S23, according to the relative importance of each modal information determined by the decision maker, the subjective weight of each modal information is obtained by the following formula: l β-1 =λ β r β Where m is the total amount of modal information, and 4. The method for real-time location recommendation based on combined weighting and user feedback according to claim 1, characterized in that: Step S3 further comprises: S31, constructing a data matrix x according to the ranking list of the location attention under the three model information calculated in step S1, where n is the total number of modal types, m is the total number of locations, and x ij Indicates the attention of the i-th location under the j-th modal information: S32, using the min-max normalization method to perform dimensionless processing on the location attention data of each modal information to obtain a normalized data matrix X; S33, calculate the standard deviation of each modal data to represent the difference fluctuation of each modal information, and the standard deviation calculation formula is as follows: Where S j is the standard deviation of the j-th modal data; S34, calculating the correlation coefficient between each modal data to indicate the degree of conflict between each modal information, the correlation coefficient calculation formula is as follows: where r jj' is the correlation coefficient between the jth modal data and the j′th modal data, σ j ,σ j' are the standard deviations of the jth modal data and the j′th modal data, μ j ,μ j' are the means of the j-th modal data and the j′-th modal data respectively; S35, then the conflict calculation formula of j modal information is as follows: S36, calculating the amount of information of each modal information: C j =S j *R j C j The larger it is, the more weight should be assigned to the jth modal information; S36, calculate the objective weight of each modal information:

5. The method for real-time location recommendation based on combined weighting and user feedback according to claim 1, characterized in that: Step S4 further comprises: S41, setting subjective weight λ j and objective weight ω j The coefficient of and ξ, the combined weights are as follows: S42, use the difference coefficient method to solve the coefficient and ξ: Where P j is the vector of subjective weights sorted in ascending order, i = 1, 2, ..., m, where m is the total amount of modal information.

6. The method for real-time location recommendation based on combined weighting and user feedback according to claim 1, characterized in that: Step S5 further comprises: S51, calculate the NDCG value of the feedback information and the sorted list of each modal information: ndcg1=NDCG(f,m) ndcg2=NDCG(f,n) ndcg3=NDCG(f,e) Where f is the feedback information, m, n and e are the place attention ranking lists under the three modes of eye movement data, operation records and text data respectively, ndcg1, ndcg2 and ndcg3 are the NDCG values ​​calculated for the feedback information and the ranking lists of each mode respectively; S52, sort the NDCG values ​​from large to small to determine the order relationship of the three modal information: sortndcg=sort(ndcg1,ndcg2,ndcg3) Among them, sort(·) sorts the NDCG values ​​from large to small, and the output sortndcg includes the NDCG values ​​represented by each modal information after sorting, which are sortndcg(0), sortndcg(1) and sortndcg(2) respectively; S53, calculate the relative importance of the three modal information: S54, calculate the relative importance of the three modal information according to the order relationship and NDCG value, and update the subjective weight.

7. A real-time location recommendation system based on combined weighting and user feedback, characterized in that: The real-time location recommendation system comprises: Data feature extraction unit: used to extract features from the location-related data in the three modes of eye movement data, operation records and text data within a period of time, and obtain a ranking list of location attention in each mode; Subjective weight generation unit: uses the G1 algorithm to determine the relative importance of the three modal information to initialize the subjective weights of the three modal information; Objective weight generation unit: Use the CRITIC algorithm to comprehensively analyze the ranking list of location attention under each mode and calculate the objective weights of the three modal information; Comprehensive weight generation unit: uses the difference coefficient method to combine the subjective weight and the objective weight to obtain the combined weight, and calculates and outputs the location recommendation list based on the combined weight; User feedback generation unit: obtains user feedback information on the place recommendation list, calculates the normalized discounted cumulative gain NDCG between it and the place attention ranking list under different modes, and updates the relative importance and subjective weight of the three modal information.

Images