Recommendation method for iot data based on user preference

Abstract

This invention relates to a recommendation method based on IoT data and user preferences, comprising the following steps: S1: Constructing an interest model of user's registered personal information and analyzing the user's registration information; S2: Analyzing user preference information based on click count, browsing time, and whether or not the data was saved, and constructing a data type model for weight and operational behavior interest analysis to obtain user preference information; S3: Analyzing the interest of neighboring users in the same item and determining the similarity of the preference information of neighboring users; S4: Analyzing the similarity of user-searched items based on search item information, and thus providing search results based on user preferences. This method introduces the idea of ​​collaborative filtering recommendation algorithm, calculates user preference information by analyzing user registration information and search operation behavior; and integrates the similarity between users and search items, enabling the search system to provide highly similar recommendation results as search feedback results based on a user's neighboring users.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) big data technology, and in particular to a recommendation method for IoT data based on user preferences. Background Art

[0002] The Internet of Things (IoT) is a new generation of heterogeneous converged networks that fully embed various sensor devices and information-related technologies into the infrastructure of various industries such as power, industry, agriculture, home, transportation, and logistics. Within this converged heterogeneous network IoT, there exist computer clusters with supercomputing, processing, and control capabilities, enabling humans to manage, control, and coordinate all users, objects, and facilities within the IoT in a more precise and dynamic manner.

[0003] As the scale of IoT applications expands, sensor devices are involved in almost every object in the world, and virtually every object can be connected to the IoT environment. Therefore, the amount of IoT data will inevitably grow exponentially. While the massive amount of IoT data enriches the resources shared by users, it also makes user retrieval a serious professional problem. The current system lacks sufficient capacity to handle such a huge amount of data, preventing users from quickly and easily finding the resources they need. This necessitates improvements to existing retrieval services to enhance their capacity.

[0004] With the development of IoT technology and the impact of heterogeneous network integration, in order to achieve flexible, secure, fast and convenient resource sharing of massive heterogeneous data in IoT, and to filter out the information needed by users from this huge amount of heterogeneous information while ensuring the independence of heterogeneous data, users need to manage and retrieve the massive heterogeneous data collected by sensor devices embedded in IoT in a unified manner. Summary of the Invention

[0005] This invention provides a recommendation method for IoT data based on user preferences, which can filter out resources that users may be interested in from massive amounts of heterogeneous IoT data according to user preferences.

[0006] To address the above problems, the technical solution adopted by this invention is as follows: a recommendation method based on user preference IoT data, comprising the following steps:

[0007] S1: Construct an interest model of user registration information and analyze user registration information;

[0008] S2: Analyze user preference information based on click count, browsing time, and whether or not the data is saved, and construct a data type model to analyze the weight and operational behavior interest, thereby obtaining user preference information;

[0009] S3: Analyze the interest of nearby users in the same project and determine the similarity of the preference information of nearby users;

[0010] S4: Analyze the similarity of user search items based on the search item information, and combine the similarity of user preference information to obtain search results.

[0011] By adopting the above technical solution, in order to enable users to find resources that meet their needs and interests in the massive heterogeneous data of the Internet of Things, the idea of ​​collaborative filtering recommendation algorithm is introduced, allowing users to rate the search items; user preference information is calculated by analyzing user registration information and search operation behavior; this technical solution fully considers the similarity between users and search items, and regards users and search items as a collaborative search network, so that the search system can provide highly similar recommendation results as search feedback results based on a user's neighboring users.

[0012] As a preferred embodiment of the present invention, the specific steps of step S1 are as follows:

[0013] S11: Construct an interest model of user registered personal information using vector space model representation;

[0014] S12: Use the TFIDF term frequency algorithm to perform term weight analysis.

[0015] As a preferred technical solution of the present invention, the specific steps of step S11 are as follows:

[0016] S111: Define the User Interest Model (UIM) for user registration information using the Vector Space Model notation, denoted as UIM = (K, W); where K represents the keywords of user registration information and W represents the weight value corresponding to the registration information.

[0017] When a search client registers, the User Interest Metric (UIM) reflects the user's true level of interest to some extent based on the personal information provided (including age, gender, occupation, preferences, etc.). UIM is represented as:

[0018] UIM = {Age, Gender, Occupation, Hobbies};

[0019] S112: According to the vector space representation method, the user registration information is represented as a 4-dimensional feature vector Hobby: Hobby={(k1,w1),(k2,w2),(k3,w3),(k4,w4)}(1);

[0020] Where k1, k2, k3, and k4 represent user age, user gender, user occupation, and user hobbies, respectively, and W1, W2, W3, and W4 represent the weight values ​​of each feature, with user age being 1, user gender being 1, user occupation being 0.8, and user hobbies being 0.5. Each dimension component consists of registration information keywords and their weight values, where the weight value indicates the importance of user registration information in a certain concept or interest. If the user has not provided personal information for a certain feature, then the weight value of that feature is set to 0.

[0021] As a preferred technical solution of the present invention, step S12 specifically includes:

[0022] S121: The formula using the TFIDF term frequency algorithm is:

[0023]

[0024] Among them, W' t'd' Indicates the frequency of term t' in document d', idf t’ It is the inverse document frequency of term t';

[0025] When using the TFIDF algorithm to statistically analyze the importance of tags in webpage source code, TF_K represents the keyword frequency vector, i.e.:

[0026]

[0027] S122: The word frequency interest degree is obtained by calculating the vector weight; then the registration information interest degree RI is obtained by formula (4), which is: RI=Hobby×TF_K(4).

[0028] As a preferred embodiment of the present invention, the specific steps of step S2 are as follows:

[0029] S21: Starting from the first search, the number of clicks a user makes on a search item is represented as follows:

[0030]

[0031] Among them, f ij The number of times user i clicked to search for item j. The total number of times user i clicked on the same type of search item;

[0032] S22: The duration of time user i browses a search item j, starting from the first search, is represented as follows:

[0033]

[0034] Among them, t ij The duration for user i to browse and search item j. The total time spent by user i browsing similar search items;

[0035] S23: Analyze whether a user saves a search item. Set S'=2 records to be saved; S'=1 records to be not saved. Combine this with the user's search item operation behavior to calculate the Operation Behavior Interest Level (OIT). The formula is:

[0036] OIT=y1×f+y2×t+y3×S' (7);

[0037] Where y1, y2, and y3 represent the weights of each operation; the weight values ​​y1, y2, and y3 are 0.33.

[0038] S24: Define a data type model Type M, and introduce the TFIDF term frequency algorithm into the weight calculation of the data type of the search item to analyze the data types that users prefer (measure the data types that users prefer); the formula for the data type model is:

[0039] Type M=〈ty_j,tyw_j〉 (8);

[0040] Where: ty_j represents all data types of the search item j, and tyw_j represents the weight value of the search item j in all data types ty_j; thus, the operational interest score (OI) of the search item j in all data types ty_j is obtained:

[0041] OI = OIT × TypeM (9);

[0042] S25: After a search is completed, the system will statistically analyze the search results provided by the user, i.e., assign a satisfaction rating to the corresponding search items. During this process, the user only needs to submit the actual rating, which is 5, 4, 3, 2, 1, or 0, where 5 represents the most interested and 0 represents the least interested. The system will then statistically analyze the user's actual rating, as shown in formula (10): a ij =1+G ij / c (10);

[0043] Among them, G ij Let c be the actual rating of user i for search item j; c is a constant used to balance the actual user ratings and avoid situations where users do not rate or rate zero, which would prevent the analysis of user preference information.

[0044] S26: Combining steps S21 to S25, obtain the formula for calculating user i's preference information for search item j:

[0045] P ij =RI ij ×OI ij ×aij (11);

[0046] Among them, RI ij OI represents the level of interest of user i in the registration information of the search item j. ij Let a represent the user i's interest level in the actions performed on search item j. ij This refers to the search results provided by the user to the search system after a single search.

[0047] As a preferred embodiment of the present invention, the specific steps of step S3 are as follows:

[0048] S31: If users x and y have both rated the same search item j, then the similarity between users x and y is calculated using the Pearson correlation coefficient. p The formula is:

[0049]

[0050] Among them, P x,j This represents the actual rating of user x for search item j, Py ,j This represents the actual rating that user y gave to search item j. This represents the average rating given by user x to all rated search items j. I represents the average rating given by user y to all rated search items j; x,y This represents the set of search items that users x and y have jointly rated;

[0051] S32: If users x and y do not have any search items with the same rating, but neither user x nor user y is a newly joined user, use cosine similarity analysis to calculate the similarity sim(x, y) between users x and y; the formula is:

[0052]

[0053] Where, ||x|| represents the vector value of user x's preference information, and ||y|| represents the vector value of user y's preference information, that is, formula (13) is transformed into the form of formula (14):

[0054]

[0055] Among them, P x,e P represents user x's actual rating of search item e; y,f This represents user y's actual rating of search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated. The smaller the cosine similarity value, the higher the similarity between user x and user y.

[0056] As a preferred embodiment of the present invention, the specific steps of step S4 are as follows:

[0057] S41: By using the user rating matrix for search items, we obtain the degree of interest among users in the search items and the similarity of user preference information. Based on user i's actual ratings of search items e and f, we obtain the similarity between search items e and f, using the following formula:

[0058]

[0059] Where e,f are the item sets I that user i has rated. i The two search items, a i,e It is user i's actual rating of search item e, a i,f It is the actual rating that user i gave to search item f. It is the average score of search item e. It is the average rating of search item f; the similarity between search items rated by user i is obtained through similarity calculation;

[0060] The cosine similarity is used to calculate the similarity between search items from different users. The formula is as follows:

[0061]

[0062] Among them, a x,e a represents the rating value of user x for search item e. y,f This represents the rating value of user y for search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated;

[0063] S42: For similarity of search items based on multiple attributes, we need to start with the similarity of each attribute. The first question to consider is whether the attributes of users x and y correspond in the two search items. However, it is not necessary to compare every attribute feature. Select data types and keywords that can serve as identifiers for search items, and use sim... key , sim Type This means that only when the similarity value of these two attributes is 1 can the two search items be judged to be extremely similar. Then, based on the score values ​​of other attributes, the attribute score similarity sim(at) is calculated using formulas (15) and (16). v This allows us to obtain the similarity between search items based on multiple attributes, using the following formula:

[0064]

[0065] Compared with existing technologies, the beneficial effects of the present invention are as follows: This recommendation method based on user preference IoT data introduces the idea of ​​collaborative filtering recommendation algorithm, requiring users to register a search client, analyzing the registration information for interest feature attributes, and weighting user interest attributes based on the importance of webpage code; during the search process, the retrieved resources are scored, and the user's search behavior is analyzed to comprehensively obtain the user's preference information; it can filter out resources that the user may be interested in from massive IoT heterogeneous data according to the user's preferences. Attached Figure Description

[0066] The technical solution of the present invention will be further described below with reference to the accompanying drawings:

[0067] Figure 1 This is a flowchart of the recommendation method based on user preference IoT data according to the present invention;

[0068] Figure 2 This is a comparison chart of the mean absolute error (MAE) of the recommendation method based on user preference IoT data according to the present invention.

[0069] Figure 3 This is a comparison chart of the root mean square error (RMSE) of the recommendation method based on user preference IoT data according to the present invention.

[0070] Figure 4 This is a comparison chart of recall rates for the recommendation method based on user preference IoT data according to the present invention. Detailed Implementation

[0071] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0072] Example: Figure 1 As shown, this recommendation method based on IoT data and user preferences includes the following steps:

[0073] S1: Construct an interest model of user registration information and analyze user registration information;

[0074] The specific steps of step S1 are as follows:

[0075] S11: Construct an interest model of user registered personal information using vector space model representation;

[0076] S111: Define the User Interest Model (UIM) for user registration information using the Vector Space Model notation, denoted as UIM = (K, W); where K represents the keywords of user registration information and W represents the weight value corresponding to the registration information.

[0077] When a search client registers, the User Interest Metric (UIM) reflects the user's true level of interest to some extent based on the personal information provided (including age, gender, occupation, preferences, etc.). UIM is represented as:

[0078] UIM = {Age, Gender, Occupation, Hobbies};

[0079] S112: Based on the vector space representation method, the user registration information is represented as a 4-dimensional feature vector. Hobby:

[0080] hobby={(k1, w1), (k2, w2), (k3, w3), (k4, w4)} (1);

[0081] Where k1, k2, k3, and k4 represent user age, user gender, user occupation, and user hobbies, respectively, and W1, W2, W3, and W4 represent the weight values ​​of each feature, with user age being 1, user gender being 1, user occupation being 0.8, and user hobbies being 0.5. Each dimension component consists of registration information keywords and their weight values, where the weight value indicates the importance of user registration information in a certain concept or interest. If no personal information is provided by the user for a certain feature, then the weight value of that feature is set to 0.

[0082] S12: Use the TFIDF term frequency algorithm to perform term weight analysis.

[0083] S121: The formula for the TFIDF term frequency algorithm is:

[0084]

[0085] Among them, W' t'd' Indicates the frequency of term t' in document d', idf t’ It is the inverse document frequency of term t';

[0086] When using the TFIDF algorithm to statistically analyze the importance of tags in webpage source code, TF_K represents the keyword frequency vector, i.e.:

[0087]

[0088] S122: The word frequency interest degree is obtained by calculating the vector weight; then the registration information interest degree RI is obtained by formula (4), which is: RI=Hobby×TF_K(4);

[0089] S2: Analyze user preference information based on click count, browsing time, and whether or not the data is saved, and construct a data type model to analyze the weight and operational behavior interest, thereby obtaining user preference information;

[0090] The specific steps of step S2 are as follows:

[0091] S21: Starting from the first search, the number of clicks a user makes on a search item is represented as follows:

[0092]

[0093] Among them, f ij The number of times user i clicked to search for item j. The total number of times user i clicked on the same type of search item k;

[0094] S22: The duration of time user i browses a search item j, starting from the first search, is represented as follows:

[0095]

[0096] Among them, t ij The duration for user i to browse and search item j. The total time spent by user i browsing similar search items k;

[0097] S23: Analyze whether a user saves a search item. Set S'=2 for saved records and S'=1 for unsaved records. Combine this with the user's search item behavior to calculate the Operation Behavior Interest Level (OIT). The formula is:

[0098] OIT=y1×f+y2×t+y3×S' (7);

[0099] Where y1, y2, and y3 represent the weights of each operation;

[0100] S24: Define a data type model TypeM, and introduce the TFIDF term frequency algorithm into the weight calculation of the data type model for retrieval item j to analyze the data types preferred by users; the formula for the data type model is:

[0101] TypeM=〈ty_j,tyw_j〉 (8);

[0102] Where: ty_j represents all data types of the search item j, and tyw_j represents the weight value of the search item j in all data types ty_j; thus, the operational interest score (OI) of the search item j in all data types ty_j is obtained:

[0103] OI = OIT × TypeM (9);

[0104] S25: After a search is completed, the system will statistically analyze the search results provided by the user, i.e., assign a satisfaction rating to the corresponding search items. During this process, the user only needs to submit the actual rating, which is 5, 4, 3, 2, 1, or 0, where 5 represents the most interested and 0 represents the least interested. The system will then statistically analyze the user's actual rating, as shown in formula (10):

[0105] a ij =1+G ij / c (10);

[0106] Among them, G ij Let c be the actual rating of user i for search item j; c is a constant used to balance the actual user ratings and avoid situations where users do not rate or rate zero, making it impossible to analyze user preference information.

[0107] S26: Combining steps S21 to S25, obtain the formula for calculating user i's preference information for search item j:

[0108] P ij =RI ij ×OI ij ×a ij (11);

[0109] Among them, RI ij OI represents the level of interest of user i in the registration information of the search item j. ij Let a represent the user i's interest level in the actions performed on search item j. ij This refers to the search results provided by the user to the search system after a single search.

[0110] S3: Analyze the interest of nearby users in the same project and determine the similarity of the preference information of nearby users;

[0111] The specific steps of step S3 are as follows:

[0112] S31: If users x and y have both rated the same search item j, then the similarity between users x and y is calculated using the Pearson correlation coefficient. p The formula is:

[0113]

[0114] Among them, P x,j This represents user x's preference information for search item j, Py ,j This represents the actual rating that user y gave to search item j. This represents the average rating given by user x to all rated search items j. I represents the average rating given by user y to all rated search items j; x,y This represents the set of search items that users x and y have jointly rated;

[0115] S32: If users x and y do not have any search items with the same rating, but neither user x nor user y is a newly joined user, use cosine similarity analysis to calculate the similarity sim(x, y) between users x and y; the formula is:

[0116]

[0117] Where ||x|| represents the vector value of user x's preference analysis information, and ||y|| represents the vector value of user y's preference analysis information, that is, formula (13) is transformed into the form of formula (14):

[0118]

[0119] Among them, P x,e P represents user x's preference information for search item e; y,f This represents user y's preference information for search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated. The smaller the cosine similarity value, the higher the similarity between user x and user y.

[0120] S4: Analyze the similarity of user search items based on the search item information, and combine this with the similarity of user preference information to obtain search results; the specific steps of step S4 are as follows:

[0121] S41: Using the user rating matrix A for the search items, obtain the degree of interest among users in the search items and the similarity of user preference information. Based on user i's actual ratings of search items e and f, obtain the similarity between search items e and f, using the following formula:

[0122]

[0123] Where e,f are the item sets I that user i has rated. i The two search items, a i,e It is user i's actual rating of search item e, a i,f It is the actual rating that user i gave to search item f. It is the average score of search item e. It is the average rating of search item f; the similarity between search items rated by user i is obtained through similarity calculation;

[0124] The cosine similarity is used to calculate the similarity between search items from different users. The formula is as follows:

[0125]

[0126] Among them, a x,e a represents the rating value of user x for search item e. y,f This represents the rating value of user y for search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated; calculate rating similarity;

[0127] S42: For similarity of search items based on multiple attributes, we need to start with the similarity of each attribute. The first question to consider is whether the attributes of users x and y correspond in the two search items. However, it is not necessary to compare every attribute feature. Select data types and keywords that can serve as identifiers for search items, and use sim... key , sim Type This means that only when the similarity value of these two attributes is 1 can the two search items be judged to be extremely similar. Then, based on the score values ​​of other attributes, the attribute score similarity sim(at) is calculated using formulas (15) and (16). v This allows us to obtain the similarity between search items based on multiple attributes, using the following formula:

[0128]

[0129] Simulation Experiment: A simulation experiment is conducted on the user information-based retrieval algorithm in the Internet of Things (IoT). This requires using Matlab R2010a to calculate the user-retrieval item rating matrix. Statistical analysis is performed on the algorithm of this technical solution and its comparison with heterogeneous data retrieval based on word segmentation technology, considering MAE, RMSE, and accuracy.

[0130] MAE (Mean Absolute Error) is used to calculate the deviation between the predicted score and the average actual score of all search items. It is used to measure the average difference between the predicted score and the actual score of search items in the Internet of Things. The smaller the MAE value, the higher the accuracy of the prediction. The MAE calculation formula is as follows (18):

[0131]

[0132] Among them, C ij The predicted rating for user i for search item j. To retrieve the actual ratings for item j, let User and I be the sets of users and the set of items to be retrieved, respectively, and let n be the number of item ratings to be retrieved.

[0133] RMSE (Root Mean Square Error) is used to measure the dispersion of search item ratings. It calculates the deviation between the predicted rating and the actual rating of the search item. Similar to MAE, the smaller the value, the higher the prediction accuracy of the algorithm. The formula is shown in (19).

[0134]

[0135] Recall, calculated by analyzing the similarity of user i's K nearest neighbors, yields the M nearest neighbors with a high degree of similarity to user i. i The retrieval system sorts the items by similarity and returns the N most similar items to user i; it calculates the relationship between the recommended retrieval items and all similar items to measure the comprehensiveness of the retrieval system; the recall rate is calculated as shown in (20):

[0136]

[0137] The MAE and RMSE results among K neighboring users are compared using an algorithm; the MAE comparison chart is shown below. Figure 1 As shown in the RMSE comparison chart... Figure 2 As shown.

[0138] from Figure 1 China MAE and Figure 2 The RMSE results show that heterogeneous data retrieval algorithms based on user information and those based on word segmentation technology obtain retrieval results by calculating the similarity between K neighboring users. As the number of neighboring users increases, the retrieval results tend to be similar, which means that the retrieval method based on user information has improved in meeting user needs.

[0139] The recall rate is compared by comparing the predicted scores based on user similarity with the actual scores for a given search item. Figure 3 As shown in the comparison of recall rates, the heterogeneous data retrieval algorithm based on user information better aligns with user preferences without altering the original data structure of the heterogeneous data.

[0140] For those skilled in the art, the specific embodiments are merely exemplary descriptions of the present invention. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A recommendation method based on IoT data of user preferences, characterized in that, Includes the following steps: S1: Construct an interest model of user registration information and analyze user registration information; S2: Analyze user preference information based on click count, browsing time, and whether or not the data is saved, and construct a data type model to analyze the weight and operational behavior interest, thereby obtaining user preference information; S3: Analyze the interest of nearby users in the same project and determine the similarity of the preference information of nearby users; S4: Analyze the similarity of user search items based on the search item information, and combine it with the similarity of user preferences to obtain search results; The specific steps of step S2 are as follows: S21: Starting from the first search, the number of clicks a user makes on a search item is represented as follows: Among them, f ij The number of times user i clicked to search for item j. The total number of times user i clicked on the same type of search item k; S22: The duration of time user i browses a search item j, starting from the first search, is represented as follows: Among them, t ij The duration for user i to browse and search item j. The total time spent by user i browsing similar search items k; S23: Analyze whether a user saves a search item. Set S'=2 for saved records and S'=1 for unsaved records. Combine this with the user's search item behavior to calculate the Operation Behavior Interest Level (OIT). The formula is: OIT=y1×f+y2×t+y3×S' (7); Where y1, y2, and y3 represent the weights of each operation; S24: Define a data type model TypeM, and introduce the TFIDF term frequency algorithm into the weight calculation of the data type model for retrieval item j to analyze the data types preferred by users; the formula for the data type model is: Type M=〈ty_j,tyw_j〉 (8); Where: ty_j represents all data types of the search item j, and tyw_j represents the weight value of the search item j in all data types ty_j; thus, the operational interest score (OI) of the search item j in all data types ty_j is obtained: OI = OIT × TypeM (9); S25: After a search is completed, the search results returned by the user to the search system are statistically analyzed using formula (10): a ij =1+G ij / c (10); Among them, G ij Let c be the actual rating given by user i to search item j; c is a constant. S26: Combining steps S21 to S25, obtain the formula for calculating user i's preference information for search item j: P ij =RI ij ×OI ij ×a ij (11); Among them, RI ij OI represents the level of interest of user i in the registration information of the search item j. ij Let a represent the user i's interest level in the actions performed on search item j. ij This refers to the search results provided by the user to the search system after a single search.

2. The recommendation method based on user preference IoT data according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11: Construct an interest model of user registered personal information using vector space model representation; S12: Use the TFIDF term frequency algorithm to perform term weight analysis.

3. The recommendation method based on user preference IoT data according to claim 2, characterized in that, The specific steps of step S11 are as follows: S111: Define the interest model UIM of user registration information using the vector space model representation method, denoted as UIM=(K,W); where K represents the keywords of user registration information and W represents the weight value corresponding to the registration information; When a client registers, the UIM is represented as follows, based on the personal information provided by the user: UIM = {Age, Gender, Occupation, Hobbies}; S112: Based on the vector space representation method, the user registration information is represented as a 4-dimensional feature vector. Hobby: Hobby={(k1,w1),(k2,w2),(k3,w3),(k4,w4)} (1); Where k1, k2, k3, and k4 represent user age, user gender, user occupation, and user hobbies, respectively, and W1, W2, W3, and W4 represent the weight values ​​of each feature.

4. The recommendation method based on user preference IoT data according to claim 3, characterized in that, Step S12 specifically includes: S121: The formula for the TFIDF term frequency algorithm is: Among them, W' t'd' Indicates the frequency of term t' in document d', idf t’ It is the inverse document frequency of term t'; When using the TFIDF algorithm to statistically analyze the importance of tags in webpage source code, TF_K represents the keyword frequency vector, i.e.: S122: The word frequency interest degree is obtained by calculating the vector weight; then the registration information interest degree RI is obtained by formula (4), which is: RI=Hobby×TF_K (4).

5. The recommendation method based on user preference IoT data according to claim 1, characterized in that, The specific steps of step S3 are as follows: S31: If users x and y have both rated the same search item j, then the similarity between users x and y is calculated using the Pearson correlation coefficient. p The formula is: Among them, P x,j This represents the actual rating of user x for search item j, Py ,j This represents the actual rating that user y gave to search item j. This represents the average rating given by user x to all rated search items j. I represents the average rating given by user y to all rated search items j; x,y This represents the set of search items that users x and y have jointly rated; S32: If users x and y do not have any search items with the same rating, but neither user x nor user y is a newly added user, use cosine similarity analysis to calculate the similarity sim(x, y) between users x and y; the formula is: Where ||x|| represents the vector value of user x's preference analysis information, and ||y|| represents the vector value of user y's preference analysis information, that is, formula (13) is transformed into the form of formula (14): Among them, P x,e P represents user x's actual rating of search item e; y,f This represents user y's actual rating of search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated.

6. The recommendation method based on user preference IoT data according to claim 5, characterized in that, The specific steps of step S4 are as follows: S41: By using the user rating matrix for search items, we obtain the degree of interest among users in the search items and the similarity of user preference information. Based on user i's actual ratings of search items e and f, we obtain the similarity between search items e and f, using the following formula: Where e,f are the item sets I that user i has rated. i The two search items, a i,e It is user i's actual rating of search item e, a i,f It is the actual rating that user i gave to search item f. It is the average score of search item e. It is the average score of the search item f; The cosine similarity is used to calculate the similarity between search items from different users. The formula is as follows: Among them, a x,e a represents the rating value of user x for search item e. y,f This represents the rating value of user y for search item f; I x I represents the set of search items that user x has rated; y This represents the set of search items that user y has rated; S42: Based on the similarity of search items with multiple attributes, the attribute score similarity sim(at) is calculated using formulas (15) and (16). v This allows us to obtain the similarity between search items based on multiple attributes, using the following formula:

Images