A method for constructing a water conservancy object metadata recommendation system based on user preferences

Abstract

The present invention discloses a method for constructing a water conservancy object metadata recommendation system based on user preferences. First, water conservancy object metadata classification labels are constructed, and index library tables are established for classification labels; secondly, user behavior logs are constructed to record user behaviors when using a retrieval system and collect user preferences; then, when users search, a retrieval depth tree is constructed to obtain the shortest search path, and facets with high discrimination are obtained. At the same time, facets with the highest correlation are recommended based on the facets selected by the user; based on the facet recommendation, similarities between different users are calculated based on the user's retrieval behavior logs for water conservancy object classification labels, and water conservancy object classification labels that are of interest to users similar to the current user are recommended; finally, the refined result set and the recommended classification labels and facets are displayed. The present invention realizes preference mining for users when using a retrieval system, and recommends search content that may be of interest to users, thereby improving the user's search efficiency.

Description

Technical Field

[0001] The invention belongs to the field of computer software, and specifically relates to a method for constructing a water conservancy object metadata recommendation system based on user preferences. Background Art

[0002] User preference refers to the bias of users in their choices during Internet browsing. Based on the preferences of different users, content that meets their needs can be recommended to different users. At present, recommendation systems based on user preferences have been well applied in the field of commercial social interaction. The user's browsing behavior or search behavior is recorded in the system log file, from which the user's characteristics are extracted and a user portrait is formed. According to the specific user portrait, personalized content can be recommended to the user. How to construct an accurate user profile and further improve the accuracy of the recommendation is the main indicator for evaluating the recommendation algorithm.

[0003] In recent years, researchers at home and abroad have conducted relevant research on faceted recommendation and proposed many faceted recommendation mechanisms. Among them, Eyal Oren et al. proposed a faceted search recommendation method based on information concepts in Daniel Tunkelang. Faceted Search: Synthesis Lectures on Information Concepts. This algorithm is limited by the scale of the information space and does not make full use of metadata, thus losing the semantic information in the metadata. At the same time, someone else proposed a persistent semantic storage method. The author used a statistical method to select keywords and selected a hierarchical clustering method to examine the guiding ability of attributes. For faceted recommendation, in addition to the two methods mentioned above, there are also methods based on user questions, user click-through rate and viewing, user rating, frequency and coverage, and XML semantic relevance. However, in semi-structured XML data, facets are relevant, and most of the above methods do not take this into account. Their recommendation methods only consider the guiding ability of a single facet.

[0004] Traditional data retrieval is achieved through keyword search metadata. Ordinary users often find it difficult to express the professional vocabulary of the water conservancy industry and cannot accurately query their needs. Among them, the object metadata in the water conservancy field is very different from the information metadata in general fields. The intelligent recommendation algorithms in the business and social fields cannot be directly applied to the water conservancy field. The number of water conservancy management objects is large, the types are diverse, the spatial distribution is wide, and the operating environment is complex. It has natural geographical dispersion. Due to the different collection methods and management differences, it has its own characteristics: First, water conservancy data has particularity: First, the amount of water conservancy data is large. The water conservancy industry department generates a large amount of data every day. With the progress of business work, the data of the water conservancy and water affairs industry is growing rapidly. For example, searching for the word "Three Gorges" in the water conservancy object metadata retrieval system will result in thousands of entries, which is difficult for users to retrieve. Second, water conservancy data is highly professional. For the establishment of data, whether it is its classification label or retrieval facet, it is not like general data, which has strong professionalism; third, the structure of water conservancy data is complex and heterogeneous. There are many different methods for collecting and processing water conservancy data. Based on this feature, semi-structured XML data is generally used as a construction carrier when storing related data. Secondly, user retrieval of water conservancy data has its own particularity: first, users conduct exploratory retrieval during the retrieval process. Users' familiarity with water conservancy knowledge varies. Due to the complex relationships of water conservancy data, users who use retrieval sometimes do not know their needs. Second, users' hierarchical retrieval. Based on the hierarchical structure of water conservancy data, users also have hierarchical characteristics during the retrieval process.

[0005] Therefore, according to the characteristics of water conservancy object metadata and users' retrieval preferences, it is necessary to recommend different fine-grained retrieval aspects so that users can integrate their retrieval needs more quickly and accurately and find the desired water conservancy data retrieval targets. Summary of the invention

[0006] Purpose of the invention: In order to overcome the shortcomings of the prior art, the present invention provides a method for constructing a water conservancy object metadata recommendation system based on user preferences, constructs an intelligent recommendation water conservancy object metadata retrieval system, and gradually recommends relevant faceted searches to users. As the search depth increases, users can locate their needs more accurately, thereby helping users improve query efficiency.

[0007] Technical solution: The present invention provides a method for constructing a water conservancy object metadata recommendation system based on user preferences, comprising the following steps:

[0008] (1) Construct metadata classification labels for water conservancy objects and create index tables for the classification labels;

[0009] (2) Construct a user behavior log and use Web data embedding technology to record user behavior when using the search system and collect user preferences;

[0010] (3) Constructing a facet recommendation module: When searching, the user selects a certain attribute of the metadata of the water conservancy object as a screening condition. The selected attribute is the facet, and the specific value of the selected attribute is the facet value: For the specific search facets established for the metadata of the water conservancy object, each time the user searches, a search depth tree is constructed based on the initial result set obtained, and all facets on the minimum depth path tree are selected to achieve a more obvious data differentiation effect when the number of facet searches is small; at the same time, the relevance of the facets is pre-calculated, and the facets with high relevance are recommended to the user, and the result set is refined according to the user's facet selection results;

[0011] (4) Construct a classification label recommendation module: mine users’ preferences for different classification labels and use them as recommendations; obtain the intersection and union of different users’ retrieval behaviors based on the users’ retrieval behavior logs for different water conservancy object classification labels, calculate the similarity between different users, and recommend water conservancy object classification labels that similar users are interested in to the current user;

[0012] (5) Build a result display module: display the refined result set and recommended classification labels and facets to users.

[0013] Furthermore, the step (1) comprises the following steps:

[0014] (11) For the metadata of water conservancy objects, four primary classification labels are divided, including rivers, lakes, water conservancy projects, monitoring sites, and others;

[0015] (12) Subdividing the primary classification labels into secondary classification labels, and recommending more accurate secondary classification labels to users in subsequent classification label recommendations;

[0016] (13) For all classification labels, a database table is constructed to store them, and they are linked to each other through foreign key indexes to establish the relationship between water conservancy object metadata and classification labels.

[0017] Furthermore, the step (2) comprises the following steps:

[0018] (21) The user generates a search request;

[0019] (22) extracting the section information retrieved by the user;

[0020] (23) Determine whether the user's search behavior meets the selection conditions of the entry point. If so, execute (24). Otherwise, take the user's click on the tag object and the search request as the initiation condition, and record the user's click times and stay time as the main behavior evaluation indicators;

[0021] (24) Create a proxy object to record the user's search behavior and store the number of clicks on each category tag and the search duration in the user behavior log table.

[0022] Furthermore, the specific search facets described in step (3) include water conservancy object classification labels, publishing units, contacts, metadata publishing time, and review time.

[0023] Furthermore, the step (3) comprises the following steps:

[0024] (31) Each time the user performs a preliminary search, a search depth tree is constructed based on the preliminary result set obtained by the search, the number of all attributes in the preliminary result set, that is, the number of facets, is calculated, and nodes are created with different attributes as roots;

[0025] (32) The remaining attributes are used as child nodes of the root node, and the preliminary result set is divided into several refined result sets according to different attribute selections;

[0026] (33) Repeat steps (31) and (32) on the child node, using the attributes that have not appeared in the depth path tree of the current node as the child node, and continue to build the retrieval depth tree until the specific data record is retrieved;

[0027] (34) The search path with the lowest depth in the tree is the path recommended for the user to choose. The attribute of each node on the path is the recommended facet, and the attribute value is the recommended facet value.

[0028] (35) Calculate the correlation scores between all recommended facets and the facet selection of the user in the previous step, sort all recommended facets according to the correlation scores, and then recommend them to the user. The correlation score calculation formula is as follows:

[0029]

[0030] Among them, ans represents the correlation between facet a and facet b, S a represents the number of search results under facet a, S b represents the number of search results under facet b, S a S b It indicates the number of search results under both facets a and b.

[0031] Furthermore, the step (4) comprises the following steps:

[0032] (41) Query the current user's search behavior operation log, poll the user's behavior operation log for each classification tag, calculate the user's preference score for each classification tag according to the preference score formula, and record the current user's preference score for each tag. The specific score calculation formula is as follows:

[0033]

[0034] Among them, x is the number of requests, y is the length of stay, and linear regression is performed on x and y respectively, and then the sum is calculated and then linear regression is performed on the sum result. The result range is between [0,1];

[0035] (42) A threshold is set. If the preference score is higher than the threshold, it is considered that the user prefers the tag. If the metadata in the preliminary result set contains the user's preferred classification tag, the tag is directly recommended to the user. If all the classification tags in the preliminary result set have never been retrieved by the user, the results are divided according to the user's preferred classification tags, and the similarity between the current user and other users is calculated using the cosine similarity formula.

[0036] (43) Based on the user similarity calculation results, select the preferred tags of k neighboring users as the recommendation results for the current user, and further calculate the current user's preference scores for these classification tags:

[0037]

[0038] Among them, b ui refers to the basic score of user u on label i, k refers to the k parameter in the KNN algorithm, N i is the set of all users who have taken actions on label i, and v is N i A single user in r ui is the final preference score of user u for label i, sim(u,v) is the user similarity between user u and user v;

[0039] (44) Predict label scores and generate metadata classification labels for recommended water conservancy objects.

[0040] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention realizes the system construction of a water conservancy object metadata recommendation system based on user preferences, combining a faceted recommendation algorithm based on relevance and a collaborative filtering algorithm based on users; the present invention realizes the retrieval preferences of different users, accurately recommends content that may be of interest to users, helps users conduct exploratory retrieval, and improves retrieval efficiency; in addition, according to the new retrieval behaviors continuously generated by users in the retrieval system, the present invention can calculate and update user preferences in real time, and provide users with real-time, accurate, and efficient user preference content recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the present invention;

[0042] Figure 2 It is a schematic diagram of the storage structure of the classification label repository table;

[0043] Figure 3 It is AOP weaving into the user behavior flow chart;

[0044] Figure 4 It is a schematic diagram of the constructed retrieval depth tree structure;

[0045] Figure 5 This is an example diagram of a retrieval depth tree established with the unit as the root node;

[0046] Figure 6 This is an example diagram of a search depth tree built with the publishing contact as the root node. DETAILED DESCRIPTION

[0047] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0048] The present invention provides a method for constructing a water conservancy object metadata recommendation system based on user preferences, such as Figure 1 As shown, the specific steps include:

[0049] Step 1: Construct water conservancy object metadata classification labels. Water conservancy object classification labels are specific retrieval facets established for the water conservancy field, and are the classification of water conservancy object metadata. They are divided into primary and secondary labels, and index tables are established for classification labels to facilitate the recommended retrieval of classification labels in subsequent steps.

[0050] Four first-level classification labels are divided for water conservancy object metadata, including rivers, lakes, water conservancy projects, monitoring sites, and others. Second-level classification labels are divided under the first-level classification labels. Among them, the first-level classification labels of rivers, lakes and reservoirs have second-level classification labels: river basins, rivers, lakes and other labels; the water conservancy project labels have reservoirs, dams, hydropower stations and other labels; the monitoring point labels have hydrological monitoring stations, soil and water conservation monitoring stations and other labels; and other labels have water resource zoning, water function zones, soil and water conservation zoning and other labels. For all classification labels, independent database tables are built for storage, and they are linked to each other through foreign key indexes to establish the relationship between water conservancy object metadata and classification labels. The database uses Oracle for storage. The specific relationship model and storage structure are as follows: Figure 2 shown.

[0051] Step 2: Build a user behavior log, using the Spring-AOP feature (i.e. aspect-oriented programming) in the Web data embedding technology to record the user's behavior when using the search system and explore the user's preferences. Figure 3 As shown, the specific steps include:

[0052] (2.1) Create an instance object, that is, the user generates a search request behavior.

[0053] (2.2) Extracting aspect information, that is, aspect information when users search.

[0054] (2.3) Determine whether the instance object meets the selection conditions of the entry point, that is, whether the user's search behavior meets the conditions. In this embodiment, the trigger condition is whether the user uses the water conservancy object classification label as a screening condition for retrieval. If it meets the conditions, execute the next step, otherwise skip directly (2.4); In this system, the user clicks on the label object and the search request as the initiation condition, and records the user's click times and stay time as the main behavior evaluation indicators.

[0055] (2.4) Create a proxy object to record the user's search behavior, and store the number of clicks and search duration of each classification tag in the user behavior log table. The table fields include: user name, user operation, response time, request method, request parameters, IP address, and creation time. At the same time, the creation time of the request is recorded, and the time difference between the two requests is used as the user's stay time when the request is encountered next time. Each click request of the user for a tag is filtered out as a user operation data set, and the subsequent user similarity will be calculated based on this data set.

[0056] Step 3: Build a facet recommendation module. When searching, users select a certain attribute of the metadata of water conservancy objects as the filtering condition. The selected attribute is the facet, and the specific value of the selected attribute is the facet value. For the specific search facets established for the metadata of water conservancy objects, each time the user searches, a search depth tree is built based on the initial result set obtained, such as Figure 4 As shown in the figure, the non-leaf nodes of the tree are certain attributes, the leaf nodes are specific data records, and the entire search path from the root node to the leaf node is recorded. All facets on the minimum depth path tree are selected, the relevance of the facets is pre-calculated, and the facets with high relevance are recommended to the user. Finally, the result set is refined according to the user's facet selection results.

[0057] (3.1) Each time a user performs a preliminary search, a search depth tree is constructed based on the preliminary result set obtained from the search. The number of all attributes in the preliminary result set, i.e., the number of facets, is calculated, and nodes are created with different attributes as roots.

[0058] (3.2) The remaining attributes are taken as child nodes of the root node, and the preliminary result set is divided into several refined result sets based on different attribute selections.

[0059] (3.3) Repeat steps (3.1) and (3.2) on the child node, using the attributes that have not appeared in the depth path tree of the current node as the child node, and continue to build the retrieval depth tree until the specific data record is retrieved.

[0060] (3.4) The search path with the lowest depth in the tree is the path recommended for user selection. The attribute of each node on the path is the recommended facet, and the attribute value is the recommended facet value.

[0061] (3.5) Calculate the relevance scores between all recommended facets and the facet selection of the user in the previous step, sort all recommended facets according to the relevance scores, and then recommend them to the user. The relevance score calculation formula is as follows:

[0062]

[0063] Among them, ans represents the correlation between facet a and facet b, S a represents the number of search results under facet a, S b represents the number of search results under facet b, S a S b It indicates the number of search results under both facets a and b.

[0064] Take the user's search for the keyword "Three Gorges" in the water conservancy object metadata retrieval system as an example. By searching for "Three Gorges" with keywords, 17,810 water conservancy object data are obtained, which is the preliminary result set D. Calculate the number of all attributes in the preliminary result set. In this system, the attributes include five facets, namely, publishing unit, publishing contact, publishing time, review time, and water conservancy object classification label. Among them, three units have published water conservancy object metadata related to the "Three Gorges", involving five publishing contacts, including five publishing times. Establish a retrieval depth tree with the publishing unit, publishing contact, and publishing time as the root node. For example, when constructing a retrieval depth tree with the publishing unit as the root node, three publishing units are involved, namely the Information Center of the Ministry of Water Resources, the Information Center of the Changjiang River Commission, and the Hydrological Bureau of the Huaihe River Commission. Based on this, the preliminary result set D is divided into three refined result sets (D1, D2, and D3). Each refined result set can be divided according to the publishing contact and publishing time. Repeat the above steps until all attribute values ​​are divided and the retrieval locates the data records that meet the specific conditions and cannot be screened anymore. Three retrieval depth trees are obtained. Figure 5 , Figure 6In the comparison, the search depth tree established with the unit and the publishing contact as the root node is shown. It can be seen that since each publishing contact belongs to a unique unit, once the publishing contact is determined, the unit is also determined, and the search path is shorter, so the facets on this path (i.e., publishing contact, publishing time) are recommended to the user. For example, when a user first uses the system, he may need to filter data according to the facet selection order of "Ministry of Water Resources → A → 2018-3-14", but when the user's search purpose is clear and he knows that A belongs to the Ministry of Water Resources, he can directly filter data according to the recommended order "A → 2018-3-14", reducing the number of facet searches. If the user selects "Unit: Ministry of Water Resources" as the facet search condition, the relevance scores of the remaining two facets (i.e., publishing contact, publishing time) and the currently selected facet need to be calculated according to the relevance score formula, and the facets are recommended according to the relevance score sorting.

[0065] Step 4: Construct a classification label recommendation module. The classification label of water conservancy objects is a specific search facet established for the water conservancy field. When users use the water conservancy object metadata retrieval system for retrieval, the classification label of water conservancy objects is the most commonly used screening condition. Therefore, mining the user's preference for different classification labels and using them as recommendations can effectively improve the user's retrieval efficiency. According to the user's retrieval behavior log, the intersection and union of the retrieval behaviors of different users are obtained, and the similarity between different users is calculated to recommend the content that similar users are interested in to the current user.

[0066] Query the current user's search behavior operation log, poll the user's behavior operation log for each classification tag, calculate the user's preference score for each classification tag according to the preference score formula, and record the current user's preference score for each tag. The specific score calculation formula is as follows:

[0067]

[0068] Among them, x is the number of requests, y is the length of stay, and linear regression is performed on x and y respectively, and then the sum is calculated and linear regression is performed on the result of the sum. The result range is between [0,1].

[0069] A threshold is set. If the preference score is higher than the threshold, it is considered that the user prefers the tag. If the metadata in the preliminary result set contains the user's preferred classification tag, the tag is directly recommended to the user. If all the classification tags in the preliminary result set have never been retrieved by the user, the results are divided according to the user's preferred classification tags, and the similarity between the current user and other users is calculated using the cosine similarity formula.

[0070] According to the user similarity calculation results, select the favorite tags of k neighboring users as the recommendation results for the current user, and further calculate the current user's preference scores for these classification tags:

[0071]

[0072] Among them, b ui refers to the basic score of user u on label i, k refers to the k parameter in the KNN algorithm, N i is the set of all users who have taken actions on label i, and v is N i A single user in r ui is the final preference score of user u for label i, and sim(u,v) is the user similarity between user u and user v. Predict label scores and generate recommended water conservancy object metadata classification labels.

[0073] Step 5: Build a result display module to show the final refined result set and recommended classification labels and facets to users.

[0074] The displayed content includes the refined result set after further search by the user, recommended classification tags, recommended search facets, and is sorted and displayed according to the user's preferred rating value. The recommended results can help users improve their search efficiency.

[0075] The present invention realizes a water conservancy object metadata recommendation system based on user preferences. The system combines a faceted recommendation algorithm based on relevance and a collaborative filtering algorithm based on user similarity to recommend more relevant and interesting content to users, thereby improving user retrieval efficiency.

Claims

1. A method for constructing a water conservancy object metadata recommendation system based on user preferences, characterized in that: The following steps are involved: (1) Construct metadata classification labels for water conservancy objects and create index tables for the classification labels; (2) Construct a user behavior log and use Web data embedding technology to record user behavior when using the search system and collect user preferences; (3) Constructing a facet recommendation module: When searching, the user selects a certain attribute of the metadata of the water conservancy object as a screening condition. The selected attribute is the facet, and the specific value of the selected attribute is the facet value: For the preset search facets established for the metadata of the water conservancy object, each time the user searches, a search depth tree is constructed based on the initial result set obtained, and all facets on the minimum depth path tree are selected; at the same time, the relevance of the facets is pre-calculated, and the facets with high relevance are recommended to the user, and the result set is refined according to the user's facet selection results; (4) Construct a classification label recommendation module: mine users’ preferences for different classification labels and use them as recommendations; obtain the intersection and union of different users’ retrieval behaviors based on the users’ retrieval behavior logs for different water conservancy object classification labels, calculate the similarity between different users, and recommend water conservancy object classification labels that similar users are interested in to the current user; (5) Build a result display module: display the refined result set and recommended classification labels and facets to users; The step (3) comprises the following steps: (31) Each time the user performs a preliminary search, a search depth tree is constructed based on the preliminary result set obtained by the search, the number of all attributes in the preliminary result set, that is, the number of facets, is calculated, and nodes are created with different attributes as roots; (32) The remaining attributes are used as child nodes of the root node, and the preliminary result set is divided into several refined result sets according to different attribute selections; (33) Repeat steps (31) and (32) on the child node, using the attributes that have not appeared in the depth path tree of the current node as the child node, and continue to build the retrieval depth tree until the specific data record is retrieved; (34) The search path with the lowest depth in the tree is the path recommended for the user to choose. The attribute of each node on the path is the recommended facet, and the attribute value is the recommended facet value. (35) Calculate the correlation scores between all recommended facets and the facet selection of the user in the previous step, sort all recommended facets according to the correlation scores, and then recommend them to the user. The correlation score calculation formula is as follows: Among them, ans represents the correlation between facet a and facet b, S a represents the number of search results under facet a, S b represents the number of search results under facet b, S a S b It indicates the number of search results under both facets a and b; The step (4) comprises the following steps: (41) Query the current user's search behavior operation log, poll the user's behavior operation log for each classification tag, calculate the user's preference score for each classification tag according to the preference score formula, and record the current user's preference score for each tag. The specific score calculation formula is as follows: Among them, x is the number of requests, y is the length of stay, and linear regression is performed on x and y respectively, and then the sum is calculated and then linear regression is performed on the sum result. The result range is between [0,1]; (42) A threshold is set. If the preference score is higher than the threshold, it is considered that the user prefers the tag. If the metadata in the preliminary result set contains the user's preferred classification tag, the tag is directly recommended to the user. If all the classification tags in the preliminary result set have never been retrieved by the user, the results are divided according to the user's preferred classification tags, and the similarity between the current user and other users is calculated using the cosine similarity formula. (43) Based on the user similarity calculation results, select the preferred tags of k neighboring users as the recommendation results for the current user, and further calculate the current user's preference scores for these classification tags: Among them, b ui refers to the basic score of user u on label i, k refers to the k parameter in the KNN algorithm, N i is the set of all users who have taken actions on label i, and v is N i A single user in r ui is the final preference score of user u for label i, sim(u,v) is the user similarity between user u and user v; (44) Predict label scores and generate metadata classification labels for recommended water conservancy objects.

2. The method for constructing a water conservancy object metadata recommendation system based on user preferences according to claim 1 is characterized in that: The step (1) comprises the following steps: (11) For the metadata of water conservancy objects, four primary classification labels are divided, including rivers, lakes, water conservancy projects, monitoring sites, and others; (12) Subdividing the primary classification labels into secondary classification labels, and recommending more accurate secondary classification labels to users in subsequent classification label recommendations; (13) For all classification labels, a database table is constructed to store them, and they are linked to each other through foreign key indexes to establish the relationship between water conservancy object metadata and classification labels.

3. The method for constructing a water conservancy object metadata recommendation system based on user preferences according to claim 1 is characterized in that: The step (2) comprises the following steps: (21) The user generates a search request; (22) extracting the section information retrieved by the user; (23) Determine whether the user's search behavior meets the selection conditions of the entry point. If so, execute (24). Otherwise, take the user's click on the tag object and the search request as the initiation condition, and record the user's click times and stay time as the main behavior evaluation indicators; (24) Create a proxy object to record the user's search behavior and store the number of clicks on each category tag and the search duration in the user behavior log table.

4. The method for constructing a water conservancy object metadata recommendation system based on user preferences according to claim 1 is characterized in that: The preset search facets described in step (3) include water conservancy object classification labels, publishing units, contacts, metadata publishing time, and review time.

Images