A knowledge graph semantic location retrieval method and device based on Skyline

Abstract

The present invention discloses a knowledge graph semantic place retrieval method and device based on Skyline. The user inputs the user's location, queries keywords, and returns the required number of places k; extracts the coordinate information contained in the place entity in the RDF graph, constructs an R‑tree index and a place keyword set, and generates a word vector set; returns the Top‑k places according to the proposed Skyline-based semantic place retrieval algorithm. The present invention proposes to use keyword fuzzy matching to replace the precise matching method in the traditional knowledge graph spatial data retrieval method, which can provide users with more potential choices; proposes a semantic place retrieval algorithm based on Skyline, which solves the extreme value problem caused by the Top‑k linear sorting function in the traditional method, achieves a balance between the two dimensions of spatial distance and semantic relevance, and improves the retrieval efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of knowledge graph spatial data retrieval, and specifically relates to a knowledge graph semantic place retrieval method and device based on Skyline. Background Art

[0002] Information retrieval is an important method and means for people to obtain information. Relevant research shows that more than 28% of the information retrieved by users in search engines is related to geographic locations, such as scenic spots, restaurants and city addresses. Traditional retrieval methods based on text keywords can no longer meet the semantic needs of users, so information retrieval based on knowledge graphs came into being. RDF (Resource Description Framework) is a data model widely used in knowledge graphs to describe entities and relationships between entities. As its data size continues to grow, some nodes have introduced spatial coordinate information, that is, RDF data contains geographic entities, which provides a new method for semantic location retrieval.

[0003] Currently, the mainstream methods for RDF data retrieval are divided into the following two types: (1) structured language (such as SPARQL) query; (2) keyword search. The structured language query method limits the access of ordinary users to RDF data and is more suitable for domain experts, while the keyword search technology increases the applicability of ordinary users. The traditional kSP (Top-k Relevant Semantic Place Retrieval) method can return a subtree with the place entity as the root node, which covers all search keywords and selects the top-k results through a ranking function. However, the following problems still exist:

[0004] (1) During the kSP search process, each keyword must be accurately matched. However, there are often a large number of words with similar meanings but different expressions between the query keyword and the RDF data keyword set. This causes the kSP query to miss a large number of potential results and ultimately results in limited results being returned;

[0005] (2) The Top-k ranking function used by kSP transforms multi-objective problems into single-objective optimization. It is insensitive to influencing factors, easily affected by weight parameters, and easily produces extreme values, resulting in query results that cannot meet user needs. Summary of the invention

[0006] Purpose of the invention: In order to overcome the deficiencies of the above-mentioned prior art, a method and device for location retrieval based on Skyline is provided, which can balance spatial distance and semantic relevance and has high retrieval efficiency.

[0007] Technical solution: The present invention provides a knowledge graph semantic location retrieval method based on Skyline, comprising the following steps:

[0008] (1) User input query: user location, query keyword, and the number of locations to be returned k;

[0009] (2) Extract the coordinate information contained in the location entity in the RDF graph, build an R-tree index and location keyword set, and generate a word vector set;

[0010] (3) Return the top-k locations based on the proposed skyline-based semantic location retrieval algorithm.

[0011] Furthermore, the process of constructing the location keyword set in step (2) is as follows:

[0012] Extract the node name from the RDF graph node URI, and use the "_" symbol together with the node label value to perform word segmentation; generate a general keyword set through deduplication processing; concatenate and dedupe the general keyword set of the place entity and its child nodes; and form a place keyword set.

[0013] Furthermore, the process of generating the word vector set in step (2) is as follows:

[0014] Input location keyword set and user query keyword set; use Word2vec model to realize word vector conversion of keywords; output word vector set.

[0015] Furthermore, the step (3) comprises the following steps:

[0016] (31) Search the R-tree based on the user's location and return a series of locations from near to far;

[0017] (32) Calculate the spatial distance between each location and the user's location;

[0018] (33) Calculate the semantic relevance between each location and the query keyword;

[0019] (34) Repeat steps (32) and (33) until all locations are traversed;

[0020] (35) The semantic location retrieval algorithm based on Skyline selects the top-k locations.

[0021] Furthermore, the step (33) comprises the following steps:

[0022] (331) Calculate the semantic similarity between the query keyword set q.ψ and the location keyword set np.ψ: For each query keyword t i , calculate t iFor each keyword k in the location keyword set j The similarity between the words S(t i ,k j ), and select the maximum value; calculate the average value of the maximum word meaning similarity as the semantic similarity S(np) between the query keyword set q.ψ and the location keyword set np.ψ;

[0023] (332) Accumulate the shortest distance between the keyword with the largest semantic similarity and the root node to obtain the looseness L(np) of the semantic location;

[0024] (333) According to the semantic relevance calculation formula, the semantic relevance between the location and the user query is calculated:

[0025]

[0026] Among them, α is the weight parameter.

[0027] Furthermore, the step (35) comprises the following steps:

[0028] (351) Calculate the degree value of each point, sort the data according to the degree value, and obtain the candidate set. The degree value calculation formula is as follows:

[0029]

[0030] in:

[0031]

[0032]

[0033] Among them, λ i,j Indicates p i The degree value of the jth dimension in the d-dimensional data set, p i [j] indicates p i The value of the j-th dimension, μ j represents the mean of the jth dimension;

[0034] (352) Use the BNL algorithm to perform skyline screening on the locations in the candidate set;

[0035] (353) Select the top-k locations based on the degree value sorting.

[0036] Based on the same inventive concept, the present invention also provides a Skyline-based knowledge graph semantic place retrieval device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the Skyline-based knowledge graph semantic place retrieval method is implemented.

[0037] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The present invention aims at achieving accurate matching of each keyword in the kSP search process, but does not take into account that in actual search scenarios, there are often a large number of words with similar meanings but different expressions between the query keyword and the RDF data keyword set, which leads to the problem that some results that meet the user's intention are omitted and returned. Fuzzy matching is used instead of accurate matching to provide users with more choices;

[0039] 2. The present invention aims at the problem that the Top-k ranking function in kSP retrieval converts the multi-objective problem into a single-objective optimization, is insensitive to influencing factors, is easily affected by weight parameters, and easily produces extreme values, resulting in the query results failing to meet user needs. A location retrieval method based on Skyline is proposed, which adopts a multi-objective optimization algorithm to solve the original multi-objective problem and finally returns the top k parallel and uncontrolled solutions. While improving the efficiency of location retrieval, the two influencing factors of spatial distance and semantic relevance are balanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the present invention;

[0041] Figure 2 This is an example graph of the RDF graph and location keyword set. DETAILED DESCRIPTION

[0042] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0043] The present invention proposes a knowledge graph semantic location retrieval method based on Skyline. Figure 1 As shown, the specific steps include:

[0044] Step 1: The user enters a query: the user's location, the query keyword, and the number of locations k to be returned.

[0045] The user enters the query: the user's location coordinates: (32.911, 35.6311), a set of query keywords: {peter, fisherman, canonised}, and the number of locations to be returned k: 1. The general meaning of this set of query keywords can be understood as: looking for a location related to a person named Peter, who is a fisherman and has been canonised.

[0046] Step 2: Preprocess the RDF data, extract the coordinate information contained in the location entity in the RDF graph, build an R-tree index and location keyword set, and generate a word vector set.

[0047] Extract the coordinate information contained in the place entity in the RDF graph and build an R-tree index; extract the node name from the RDF graph node URI, and use the "_" symbol together with the node label value to perform word segmentation, generate a general keyword set through deduplication processing, and concatenate and dedupe the general keyword set of the place entity and its child nodes to form a place keyword set. Figure 2 (a) shows the entities extracted from the knowledge graph. <bethsaida>RDF graph description, Figure 2 (b) shows the entity <bethsaida>The general keyword set of Figure 2 (c) is the entity <bethsaida>Input the location keyword set, use the Word2vec model to convert the keywords into word vectors, and output the word vector set.

[0048] Step 3: Return the top-k locations according to the proposed skyline-based semantic location retrieval algorithm.

[0049] (1) Search the R-tree based on the user's location, return a series of locations from near to far, and traverse these locations in order. <bethsaida>Take as an example to show the construction process of semantic place.

[0050] (2) Calculation <bethsaida>The spatial distance from the user's location, <bethsaida>The coordinates are (32.91, 35.631), D = 0.111km.

[0051] (3) Calculate the query keyword set q.ψ and <bethsaida>The semantic similarity S(np) and the looseness L(T np ). For the query keyword peter, <bethsaida>The word with the highest similarity in the location keyword set is peter, with a similarity of 1 and a shortest distance from the root node np of 1; for the query keyword fisherman, <bethsaida>The word with the highest similarity in the location keyword set is fishing, with a similarity of 0.86 and a shortest distance from the root node np of 0; for the query keyword canonised, <bethsaida>The word with the highest similarity in the location keyword set is saint, with a similarity of 0.81, and the shortest distance from the root node np is 1. Therefore, the semantic similarity S(np) = (1+0.86+0.81) / 3 = 0.89, and the looseness L(T np )=1+1+0+1=3. The area obtained by query is <bethsaida>, has the tag [Fishing_villages], is<Saint_Peter> 's birthplace, with the label [Anglican_saints], meets the query expectation.

[0052] (4) According to the semantic relevance calculation formula, calculate <bethsaida>The semantic relevance to the user query. The semantic relevance calculation formula is as follows:

[0053]

[0054] Among them, α is a weight parameter, which is set to 0.5 in this example, and the semantic relevance f(np)=(0.89*0.5) / (3*0.5)=0.296.

[0055] (5) Repeat steps (2) to (4) until all locations have been traversed.

[0056] (6) Calculate the degree value of each location, sort the data according to the degree value, obtain the candidate set, and use the BNL algorithm to perform skyline screening on the locations in the candidate set; select the top-1 location based on the degree value sorting.

[0057] The degree value calculation formula is as follows:

[0058]

[0059] in,

[0060]

[0061]

[0062] Among them, λ i,j Indicates p i The degree value of the jth dimension in the d-dimensional data set, p i [j] indicates p i The value of the j-th dimension, μ j Represents the mean of the j-th dimension.

[0063] In this example, p1[1] means <bethsaida>The spatial distance from the query location, p1[2] represents <bethsaida>The semantic relevance to the query keyword is p1[1]=0.111, p1[2]=0.296, μ1 represents the mean spatial distance, μ2 represents the mean semantic relevance, μ1=0.156,, μ2=0.194, then φ(p1)=0.237.

[0064] Based on the same inventive concept, the present invention also provides a Skyline-based knowledge graph semantic place retrieval device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the Skyline-based knowledge graph semantic place retrieval method is implemented.< / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida> < / bethsaida>

Claims

1. A knowledge graph semantic location retrieval method based on Skyline, characterized in that: The following steps are involved: (1) User input query: user location, query keyword, and the number of locations to be returned k; (2) Extract the coordinate information contained in the location entity in the RDF graph, build an R-tree index and location keyword set, and generate a word vector set; (3) Return the top-k locations according to the proposed skyline-based semantic location retrieval algorithm; The step (3) comprises the following steps: (31) Search the R-tree based on the user's location and return a series of locations from near to far; (32) Calculate the spatial distance between each location and the user's location; (33) Calculate the semantic relevance between each location and the query keyword; (34) Repeat steps (32) and (33) until all locations are traversed; (35) The semantic location retrieval algorithm based on Skyline selects the top-k locations; The step (35) comprises the following steps: (351) Calculate the degree value of each point, sort the data according to the degree value, and obtain the candidate set. The degree value calculation formula is as follows: in: Among them, λ i,j Indicates p i The degree value of the jth dimension in the d-dimensional data set, p i [j] indicates p i The value of the j-th dimension, μ j represents the mean of the jth dimension; (352) Use the BNL algorithm to perform skyline screening on the locations in the candidate set; (353) Select the top-k locations based on the degree value sorting.

2. The Skyline-based knowledge graph semantic location retrieval method according to claim 1, characterized in that: The implementation process of constructing the location keyword set in step (2) is as follows: Extract the node name from the RDF graph node URI, and use the "_" symbol together with the node label value to perform word segmentation; generate a general keyword set through deduplication processing; concatenate and dedupe the general keyword set of the place entity and its child nodes; and form a place keyword set.

3. The Skyline-based knowledge graph semantic location retrieval method according to claim 1, characterized in that: The implementation process of generating the word vector set in step (2) is as follows: Input location keyword set and user query keyword set; use Word2vec model to realize word vector conversion of keywords; output word vector set.

4. The Skyline-based knowledge graph semantic location retrieval method according to claim 1, characterized in that: The step (33) comprises the following steps: (331) Calculate the semantic similarity between the query keyword set q.ψ and the location keyword set np.ψ: For each query keyword t i , calculate t i For each keyword k in the location keyword set j The similarity between the words S(t i ,k j ) and select the maximum value; Calculate the average value of the maximum word meaning similarity as the semantic similarity S(np) between the query keyword set q.ψ and the location keyword set np.ψ; (332) The shortest distance between the keyword with the largest semantic similarity and the root node is accumulated to obtain the looseness L(T np ); (333) According to the semantic relevance calculation formula, the semantic relevance between the location and the user query is calculated: Among them, α is the weight parameter.

5. A knowledge graph semantic location retrieval device based on Skyline, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the Skyline-based knowledge graph semantic place retrieval method according to any one of claims 1 to 4 is implemented.

Images