Hotel matching method based on multidimensional relation reasoning

Abstract

The invention discloses a hotel matching method based on multi-dimensional relation reasoning, and the method comprises the steps: obtaining to-be-matched hotel information, carrying out the preprocessing of the to-be-matched hotel information, and generating effective to-be-matched information; obtaining target hotel information, wherein the target hotel information and the effective to-be-matched information comprise name information and position information; loading a hotel matching knowledge base, and calculating the comprehensive confidence of the target hotel information and the effective to-be-matched information; generating a matching result intelligent decision according to the comprehensive confidence coefficient; the matching result intelligent decision comprises matching success and manual auditing; wherein the hotel matching knowledge base adopts a structured storage hierarchical design to manage the matching relationship of hotels, and provides a calculation model for calculating the comprehensive confidence coefficient and a data cache space. According to the technical scheme, the matching accuracy of the hotel information can be effectively improved; and a hotel information standardization solution which is high in reliability, low in cost and easy to expand is provided for the tourism industry.

Description

Technical Field

[0001] This invention relates to the fields of big data analysis and artificial intelligence technology, and more specifically, to a hotel matching method based on multidimensional relational reasoning. Background Technology

[0002] In online travel and travel management scenarios, tourists initiate hotel searches to navigation systems and hotel management platforms through various methods such as voice assistants, map navigation, or manual input. These systems, in turn, use their own data to search for hotel information with OTAs (Online Travel Agencies). However, inconsistencies often arise between the hotel information provided by tourists, navigation systems, hotel management platforms, and OTAs. For example, tourists may input colloquial hotel names, inaccurate addresses, or text with recognition errors; the hotel information on hotel management platforms may differ from the hotel database registered with OTAs; and changes to hotel names may not be synchronized. Therefore, search results often do not match user needs.

[0003] To address this issue, existing technologies often employ matching methods based on single-dimensional features (such as name or address). This approach lacks the comprehensive analytical capabilities to handle complex business relationships, user behavior patterns, and historical dynamic data, resulting in low matching accuracy (approximately 60%-70%) and an inability to effectively handle fuzzy queries and cross-regional chain brand identification. While some multi-dimensional relationship modeling frameworks have emerged, their application in hotel matching relies on fixed weights and lacks the self-learning capabilities of a knowledge base, leading to low accuracy in cases of address changes, name changes, and aliases.

[0004] Therefore, there is an urgent need for a hotel matching method that can achieve efficient hotel information matching through a continuously learning hotel relationship model, addressing complex situations such as inconsistencies in hotel information from multiple parties, cross-regional issues, and the need to identify chain brands. Summary of the Invention

[0005] To achieve the above objectives, this application provides a hotel matching method based on multidimensional relational reasoning, comprising the following steps: Obtain hotel information to be matched, preprocess the hotel information to be matched, and generate valid matching information; Obtain target hotel information, which, along with the valid matching information, includes name information and location information. Load the hotel matching knowledge base and calculate the combined confidence level of the target hotel information and the valid match information; Based on the overall confidence level, an intelligent decision is generated for the matching results; the intelligent decision includes successful matching and manual review. The hotel matching knowledge base employs a structured, layered storage design to manage hotel matching relationships, providing a calculation model and data caching space for calculating overall confidence. To manage hotel matching relationships, the hotel matching knowledge base includes a relationship definition management library, a historical matching case library, a user feedback knowledge base, and a hotel relationship library. To calculate overall confidence, the hotel matching knowledge base provides a relationship strength calculator and a dynamic weight calculator.

[0006] The relationship definition management library is used to pre-set matching rules. The matching rules are realized through the dimensional relationships between the target hotel information and the valid information to be matched. The dimensional relationships include: semantic relationships, spatial relationships, business relationships and historical relationships. The historical matching case library is used to store a large number of historical matches and labeled data that reflects the matching results; The user feedback knowledge base is used to store user feedback information on matching results; The hotel relationship database manages the relationships between hotels that have participated in matching through nodes and edges. Nodes represent hotel information, edges represent the latest matching results between hotels, and multi-dimensional hotel feature relationships, relationship strength values ​​for each dimension, new weights for each dimension, and overall confidence scores are the attributes of the edges.

[0007] The calculation of the combined confidence level of the target hotel information and the valid match information includes the following steps: Perform relationship matching between target hotel information and valid matchable information to obtain relationship results in various dimensions; Quantify the relationship strength between target hotel information and valid matching information, and generate the quantified relationship strength baseStrength for each dimension; Based on the strength of the relationship in each dimension, the weights of each dimension are dynamically adjusted to generate semantic dynamic weights, spatial dynamic weights, business matching dynamic weights, and historical dynamic weights. The overall confidence level is calculated based on the weights of each dimension.

[0008] Obtaining relationship results across various dimensions includes: The BERT model is used to calculate the name similarity between two hotels, and the semantic relationship is determined as the relation result. Calculate the straight-line distance between the coordinates of the hotel information to be matched and the target hotel information, and use the straight-line distance as the relationship result; Analyze data from the enterprise credit information disclosure system to determine the business relationship between the two hotels as the relationship result; Historical matching records of the hotel information to be matched and the target hotel information are extracted from the historical matching case database as matching evidence chain data, and the historical relationship between the two hotels is determined as the relationship result.

[0009] The calculation method for the quantitative relationship strength baseStrength is as follows: baseStrength=baseStrength×(k1+k2×supportFactor+k3×timeFactor+k4×consistencyFactor), Where k1, k2, k3, and k4 are the initial weights and the weights of the support factor, time factor, and consistency factor, respectively, and supportFactor, timeFactor, and consistencyFactor are the values ​​of the support factor, time factor, and consistency factor, respectively.

[0010] Furthermore, the support factor is used to measure the degree of historical accumulation of relational evidence. The support factor is calculated as follows: supportFactor = log(1 + number of times the relation occurs) / log(100).

[0011] The time factor, based on half-life, calculates the dynamic decay of relationship strength. The calculation method is as follows: timeFactor = Date Difference / Half-life, where: Date Difference is the absolute value of the date difference between the current time and the last matching time in the matching evidence, and the half-life is usually set to 30 days.

[0012] The consistency factor is used to determine the consistency between the current match and the previous match result. It is calculated as follows: consistencyFactor = successRate × (1 - Variance); where successRate is the historical success rate and variance is the variance. Furthermore, the historical success rate is based on the ratio of the number of successful matches for that relationship within the same hotel category in the hotel matching knowledge base to the total number of matches; Variance is the degree of dispersion of multiple matching results of relational results in a specific region. It is calculated as follows: Variance = Σ(Matching results based on the relational result each time - Average of matching results based on the relational result)² / n.

[0013] Furthermore, dynamically adjusting the weights of each dimension includes the following steps: Basic weights are configured for multiple relational dimensions, with the sum of the basic weights being 1.0. The basic weights include: semantic dimension basic weight, spatial dimension basic weight, business dimension basic weight, and historical dimension basic weight. The method for dynamically adjusting the weights of each dimension is as follows: New semantic dimension weight = basic semantic dimension weight × semantic relation strength value. New spatial dimension weight = basic spatial dimension weight × spatial relationship strength value. New weight for the business dimension = basic weight for the business dimension × business relationship strength value. New weight for historical dimension = base weight for historical dimension × historical relationship strength value; The adjusted weights of each dimension are normalized to generate semantic dynamic weights, spatial dynamic weights, business matching dynamic weights, and historical dynamic weights.

[0014] The method for calculating the overall confidence level is as follows: Overall confidence score = Semantic weighted confidence score + Spatial weighted confidence score + Commercial weighted confidence score + Historical weighted confidence score Wherein, semantic weighted confidence = semantic dynamic weight × semantic relationship similarity, Spatial weighted confidence score = Spatial dynamic weight × (1 - Straight-line distance / Distance standard), Business-weighted confidence score = Business dynamic weight × Business relationship value Historical weighted confidence score = historical matching weight × (historical matching success rate of semantic relationship + historical matching success rate of spatial relationship + historical matching success rate of commercial relationship) / 3.

[0015] This invention employs a multi-dimensional feature fusion and dynamic weighting mechanism to effectively handle complex scenarios such as colloquial names, ambiguous addresses, and cross-system data differences, thereby significantly improving the accuracy of hotel information matching. Simultaneously, it automatically captures market dynamics such as hotel name changes, address changes, and brand chain relationships during the matching process, and continuously optimizes matching accuracy through a self-learning mechanism, providing the tourism industry with a highly reliable, low-cost, and easily scalable standardized hotel information solution. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a hotel matching method based on multidimensional relational reasoning according to an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a hotel matching method that constructs a hotel matching knowledge base based on multidimensional relational reasoning, adaptively and dynamically adjusts and optimizes the recognition strength, and improves the accuracy of hotel information matching through continuous application.

[0018] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] The implementation process of the hotel matching method based on multidimensional relational reasoning provided by this invention is as follows: Figure 1 As shown, it includes the following steps: Step S100: Obtain the hotel information to be matched, preprocess the hotel information to be matched, and generate valid matching information; This step obtains hotel information to be matched from the demand terminal (such as user terminal, management platform), including name information and location information.

[0020] Since obtaining hotel information for matching can be done through various methods such as manual input and voice recognition, the information may be inaccurate or incomplete. Therefore, this step involves preprocessing the hotel information, specifically including: Name information cleaning involves removing meaningless characters, standardizing the format, and extracting core keywords to support multilingual processing in subsequent steps. Location information cleaning and parsing: Standardize administrative region information and extract fine-grained information such as house number and building number from address information; The preprocessed name and location information constitute valid matching information.

[0021] Step S110: Obtain target hotel information; Based on the valid information to be matched, a search is performed in the OTA hotel database. If completely identical OTA hotel information is found, it means that the information of the hotel to be matched is completely consistent in the OTA hotel database, and no further matching operation is required. Otherwise, a fuzzy search is performed to find similar hotel information as the target hotel.

[0022] In this invention, the target hotel information and the hotel information to be matched correspond to each other, involving name information and location information.

[0023] Step S120: Load the hotel matching knowledge base and calculate the combined confidence level of the target hotel information and the valid matchable information.

[0024] The hotel matching knowledge base uses a structured, layered storage design to manage hotel matching relationships and provides a computational model and data cache space for calculating the overall confidence level.

[0025] To manage hotel matching relationships, the hotel matching knowledge base includes a relationship definition management library, a historical matching case library, a user feedback knowledge base, and a hotel relationship library.

[0026] 1) The relationship definition management library is used to pre-define matching rules; Specifically, the matching rules are implemented through the relationship between target hotel information and valid matchable information in four dimensions. The four dimensions are: Semantic relationships: Based on the hotel names in the target hotel information and valid match information, synonyms are analyzed using NLP to determine aliases (e.g., "Hilton DoubleTree" and "Hilton Garden Inn"). Semantic relationships specifically include: brand aliases, equivalent translations, name abbreviations, and description relevance. Spatial relationships: Calculate the physical distance (error < 50 meters) based on the GIS coordinates of the target hotel and valid matching information. Spatial relationships include: same building, same business district, and adjacent location; Business relationships: This involves leveraging publicly available information from target hotel data and available matching data to uncover industry chain connections such as brand affiliation and franchise agreements. Business relationships include: relationships with the same brand, chain relationships, business partnerships, and competitive relationships. Historical relationships: Based on user behavior learning, including historical matching results and manual confirmations, it is determined whether the hotel entities at both ends of the match have matching records. Historical relationships are determined through a historical matching case library and a user feedback knowledge base.

[0027] 2) The historical matching case library is used to store a large number of historical matches and labeled data that reflects the matching results; 3) The user feedback knowledge base is used to store user feedback information on the matching results; it is used to correct the hotel relationship based on the user feedback information after the matching results are generated.

[0028] In the process of calculating the overall confidence score, the hotel matching knowledge base provides a calculation model including a relationship strength calculator and a dynamic weight calculator, and provides caching of relationship strength and dynamic weight data during the calculation process; it also provides caching of historical records such as the historical matching case library during the calculation process.

[0029] 4) The hotel relationship database manages and maintains the relationships between hotels that have participated in the matching through nodes and edges. Nodes are hotel information, edges are the latest matching results between hotels, and multi-dimensional hotel feature relationships, relationship strength values ​​of each dimension, new weights of each dimension, and comprehensive confidence are the attributes of the edges.

[0030] Calculating the combined confidence level of the target hotel information and the valid match information involves the following steps: Step S121 performs relationship matching to obtain relationship results for each dimension: In this step, the relationship data between the target hotel information and the valid matchable information is calculated across four dimensions, including: 1) Use the BERT model to calculate the name similarity between the two hotels and determine the semantic relationship as the relation result; When calculating the name similarity of two hotels, Jaro-Winkler (prefix matching), cosine similarity (word vectors), and Levenshtein (edit distance) are used to calculate the name similarity of the two hotels, and the semantic relationship is determined as the relation result; When performing Jaro-Winkler (prefix matching), an edit distance algorithm that is friendly to prefix matching is used. It calculates the number of matching characters and the number of inversions between the two names, then adds a prefix score, outputting a similarity score of 0 to 1 (the closer to 1, the more similar). Suitable for hotel names (e.g., "XX Hotel" and "XX Hotel (YY Branch)"). When calculating cosine similarity, the names are segmented into word vectors and then converted into word vectors to calculate the similarity, in order to better capture the semantics; the output of the cosine similarity calculation result is 0~1, and the closer to 1, the more similar they are. When performing Levenshtein (edit distance), the degree of difference between two names is calculated: the minimum number of editing operations (insertion, deletion, replacement) required to convert the first name to the second name is calculated, and then the similarity is obtained by subtracting 1 (edit distance / maximum string length). The similarity value ranges from 0 to 1 (the closer to 1, the more similar).

[0031] Based on the above method, semantic relationship similarity is calculated and expressed as follows: Semantic relation similarity = w1×score_jw+w2×score_cos+w3×score_lev, where score_jw is the prefix matching score, score_cos is the cosine score, core_lev is the edit distance score, and w1, w2 and w3 are the weights of prefix matching, cosine and edit distance, respectively. The weights are adjusted based on the sum of the scores: when score_jw > score_cos and score_jw > score_lev: w1(Jaro-Winkler) = 0.5, w2(cosine) = 0.3, w3(Levenshtein) = 0.2; when score_cos > core_jw and score_cos > score_lev: w1(Jaro-Winkler) = 0.3, w2(cosine) = 0.4, w3(Levenshtein) = 0.3; when score_lev > core_jw and score_lev > score_cos: w1(Jaro-Winkler) = 0.3, w2(cosine) = 0.3, w3(Levenshtein) = 0.4.

[0032] 2) Obtain the latitude and longitude coordinates of the hotel information to be matched and the target hotel information through the Tianditu API or Gaode Map API respectively, and then calculate the straight-line distance between the two coordinates using the Haversine formula, and use the straight-line distance as the relationship result.

[0033] 3) Analyze the data from the enterprise credit information disclosure system to determine the business relationship between the two hotels as the relationship result.

[0034] 4) Extract historical matching records of the hotel information to be matched and the target hotel information from the historical matching case library as matching evidence chain data, and determine the historical relationship between the two hotels as the relationship result.

[0035] Step S122 quantifies the relationship strength between the target hotel information and the valid matching information, and generates the quantified relationship strength baseStrength for each dimension; The method for calculating the quantitative relationship strength (baseStrength) is as follows: baseStrength=baseStrength×(k1+k2×supportFactor+k3×timeFactor+k4×consistencyFactor), Where k1, k2, k3 and k4 are the initial weights and the weights of the support factor, time factor and consistency factor, respectively, and can be assigned values ​​of 0.35, 0.29, 0.22 and 0.14; supportFactor, timeFactor and consistencyFactor are the values ​​of the support factor, time factor and consistency factor, respectively.

[0036] The support factor measures the historical accumulation of evidence for a relationship. The support factor is calculated as: supportFactor = log(1 + number of times the relationship occurs) / log(100). The support factor ranges from [0,1], with a higher support factor value indicating a greater frequency of relationship occurrences. The time factor, based on half-life, calculates the dynamic decay of relationship strength. The calculation method is as follows: timeFactor = Date Difference / Half-Life, where: Date Difference is the absolute value of the date difference between the current time and the last match time in the matching evidence, and the half-life is usually set to 30 days; The consistency factor is used to determine the consistency between the current match and the previous match, ensuring that relationships with high success rates and low variance receive higher weights; its calculation formula is: consistencyFactor = successRate × (1 - Variance); where successRate is the historical success rate and variance is the variance.

[0037] Historical success rate is the ratio of the number of successful matches for the same hotel category (e.g., budget or luxury) to the total number of matches in the knowledge base. Variance is the degree of dispersion of the relationship results of the four dimensions in a specific area (such as a city or business district) across multiple matching results. The calculation method is: Variance = Σ(Match result based on the relationship result each time - Average of the matching results based on the relationship result)² / n; Furthermore, the consistency factor results are normalized to the [0,1] interval, with higher values ​​indicating a more stable relationship.

[0038] Through multi-factor fusion calculation, a quantitative relationship strength baseStrength is generated, which can strengthen the matching evidence chain and improve the matching confidence.

[0039] Step S123: Dynamically adjust the weights of each dimension based on the strength of the relationship between each dimension; 1) First, configure the basic weights for the four relational dimensions of hotel matching. To ensure a balanced relative importance of each dimension, the sum of the basic weights is initially set to 1.0. For example, the configuration is as follows: basic weight for semantic dimension is 0.3, basic weight for spatial dimension is 0.3, basic weight for commercial dimension is 0.15, and basic weight for historical dimension is 0.25.

[0040] 2) Based on the relationship strength corresponding to each dimension, the weights of each dimension are dynamically adjusted. The adjusted weights are as follows: New semantic dimension weight = basic semantic dimension weight (0.3) × semantic relation strength value; New weight for spatial dimension = Basic weight for spatial dimension (0.3) × Spatial relationship strength value; New weight for the business dimension = Basic weight for the business dimension (0.15) × Business relationship strength value; New weight for historical dimension = Basic weight for historical dimension (0.25) × Historical relationship strength value.

[0041] 3) Since the sum of the weights of each dimension after dynamic adjustment may not be 1, the adjusted weights of each dimension are normalized to generate semantic dynamic weights, spatial dynamic weights, business matching dynamic weights, and historical dynamic weights. The normalized weight values ​​ensure that the sum of all weights is 1.0, avoiding the influence of weight bias on the decision results.

[0042] The normalization formula is: Normalized weight = Adjusted weight / Sum of adjusted weights for each dimension. After normalization, the weight values ​​for each dimension remain within the range of [0,1] and are comparable, providing standardized input for subsequent comprehensive confidence calculation.

[0043] Through the above steps, the weight values ​​of the hotel matching dimension can be adaptively adjusted in real time according to the strength of the relationship, breaking through the limitations of traditional static weights and improving the accuracy and flexibility of matching decisions.

[0044] Step S124 calculates the overall confidence level based on the weights of each dimension.

[0045] The core of calculating the overall confidence level is to use a simplified formula based on an improved DS evidence theory to fuse multi-dimensional evidence.

[0046] First, calculate the weighted confidence scores for the four dimensions: 1) Semantic weighted confidence score = Semantic dynamic weight × Semantic relationship similarity. The semantic relationship similarity is obtained during relationship matching in step S121.

[0047] 2) Spatial weighted confidence score = Spatial dynamic weight × (1 - Straight-line distance / Distance standard), The distance standard is a pre-set standard value (such as 500 meters). Hotels that are more than the distance standard are considered to be different hotels.

[0048] 3) Business-weighted confidence score = Business dynamic weight × Business relationship value. Among them, the business relationship value is the quantification of business relationships, which can be pre-defined, including: "Same brand relationship" set to 0.9, "Same chain relationship" set to 0.65, "Business cooperation relationship" set to 0.3, "Competition relationship" set to 0.1, and "No relationship" set to 0.

[0049] 4) Historical weighted confidence = historical matching weight × (historical matching success rate of semantic relationship + historical matching success rate of spatial relationship + historical matching success rate of commercial relationship) / 3. If there is no historical relationship matching record, the historical matching success rate is 1.

[0050] Finally, the overall confidence level is calculated as follows: Overall confidence level = semantically weighted confidence level + spatially weighted confidence level + commercially weighted confidence level + historically weighted confidence level; The overall confidence level is in the range of [0,1], with a higher value indicating a closer approximation.

[0051] Step S130: Generate intelligent decision-making based on comprehensive confidence level to determine whether the match is successful; In this step, the overall confidence level is graded. If the overall confidence level is ≥0.9, the match is successful and the best match result is returned. If the overall confidence level is in the range of 0.7-0.89, the manual review process is initiated. Otherwise, it is marked as an invalid match and an exception log is recorded.

[0052] In this step, step S131 also needs to be performed to update the hotel relationships in the hotel matching knowledge base.

[0053] Step S131: Update the hotel relationships in the hotel matching knowledge base.

[0054] In this step, it is determined whether the hotel relationship database of the hotel matching knowledge base contains data with target hotel information and valid matchable information as nodes. If not, a new node is established based on the valid matchable information and / or the target hotel information, with the intelligent decision of the current matching result as the edge. The multi-dimensional hotel feature relationship between the target hotel information and the valid matchable information, the relationship strength value of each dimension, the new weight of each dimension, and the comprehensive confidence level are used as the attributes of the edge to establish the hotel relationship between the target hotel information and the valid matchable information.

[0055] If the hotel database already contains data with the target hotel information and valid matching information as nodes, and there is no edge between the two nodes, then the intelligent decision based on the matching result will establish an edge between the two nodes and define the attributes of the edge. If a hotel relationship has been established between the target hotel information and the valid matching information, the edge between the two nodes will be updated based on the matching result, and the attributes of the edge will also be updated.

[0056] At the same time, the relevant information of this match will be written into the historical match case database.

[0057] Step S140: In the manual review process, determine whether the manual review was successful; based on the result of the manual review, execute step S131 to update the hotel matching knowledge base; if the manual review indicates a matching failure, mark it as an invalid match and record the exception log.

[0058] This invention breaks through the limitations of traditional single-dimensional matching by adopting a multi-dimensional feature fusion and dynamic weighting mechanism to effectively handle complex scenarios such as colloquial names, fuzzy addresses, and cross-system data differences, thereby significantly improving the accuracy of hotel information matching. At the same time, it automatically captures market dynamics such as hotel name changes, address changes, and brand chain relationships during the matching process, and continuously optimizes the matching accuracy through a self-learning mechanism, providing the tourism industry with a highly reliable, low-cost, and easily scalable standardized hotel information solution.

[0059] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A hotel matching method based on multidimensional relational reasoning, characterized in that, Includes the following steps: Obtain hotel information to be matched, preprocess the hotel information to be matched, and generate valid matching information; Obtain target hotel information, wherein the target hotel information and the valid matching information include name information and location information; Load the hotel matching knowledge base and calculate the combined confidence level of the target hotel information and the valid match information; Based on the comprehensive confidence level, an intelligent decision is generated for the matching result; the intelligent decision includes successful matching and manual review. Among them, the hotel matching knowledge base adopts a structured storage layered design to manage the matching relationship of hotels, and provides a calculation model for calculating the comprehensive confidence score and data cache space; To manage hotel matching relationships, the hotel matching knowledge base includes a relationship definition management library, a historical matching case library, a user feedback knowledge base, and a hotel relationship library; to calculate the overall confidence score, the hotel matching knowledge base provides a relationship strength calculator and a dynamic weight calculator.

2. The hotel matching method based on multidimensional relational reasoning according to claim 1, characterized in that, The relationship definition management library is used to pre-set matching rules. The matching rules are realized through the dimensional relationships between target hotel information and valid information to be matched. The dimensional relationships include: semantic relationships, spatial relationships, business relationships and historical relationships. The historical matching case library is used to store a large number of historical matches and labeled data that reflects the matching results; The user feedback knowledge base is used to store user feedback information on the matching results; The hotel relationship database manages the relationships between hotels that have participated in matching through nodes and edges. Nodes represent hotel information, edges represent the latest matching results between hotels, and multi-dimensional hotel feature relationships, relationship strength values ​​for each dimension, new weights for each dimension, and overall confidence scores are the attributes of the edges.

3. The hotel matching method based on multidimensional relational reasoning according to claim 1, characterized in that, The calculation of the combined confidence level of the target hotel information and the valid match information includes the following steps: Perform relationship matching between target hotel information and valid matchable information to obtain relationship results in various dimensions; Quantify the relationship strength between target hotel information and valid matching information, and generate the quantified relationship strength baseStrength for each dimension; Based on the relationship strength of each dimension, the weights of each dimension are dynamically adjusted to generate semantic dynamic weights, spatial dynamic weights, business matching dynamic weights, and historical dynamic weights. The overall confidence level is calculated based on the weights of each dimension.

4. The hotel matching method based on multidimensional relational reasoning according to claim 3, characterized in that, The process of obtaining the relationship results for each dimension includes: The BERT model is used to calculate the name similarity between two hotels, and the semantic relationship is determined as the relation result. Calculate the straight-line distance between the coordinates of the hotel information to be matched and the target hotel information, and use the straight-line distance as the relationship result; Analyze data from the enterprise credit information disclosure system to determine the business relationship between the two hotels as the relationship result; Historical matching records of the hotel information to be matched and the target hotel information are extracted from the historical matching case database as matching evidence chain data, and the historical relationship between the two hotels is determined as the relationship result.

5. The hotel matching method based on multidimensional relational reasoning according to claim 3, characterized in that, The calculation method for the quantitative relationship strength baseStrength is as follows: baseStrength=baseStrength×(k1+k2×supportFactor+k3×timeFactor+k4×consistencyFactor), Where k1, k2, k3, and k4 are the initial weights and the weights of the support factor, time factor, and consistency factor, respectively, and supportFactor, timeFactor, and consistencyFactor are the values ​​of the support factor, time factor, and consistency factor, respectively.

6. The hotel matching method based on multidimensional relational reasoning according to claim 5, characterized in that, The support factor is used to measure the degree of historical accumulation of relation evidence. The support factor is calculated as follows: supportFactor = log(1 + number of times the relation occurs) / log(100).

7. The hotel matching method based on multidimensional relational reasoning according to claim 5, characterized in that, The time factor, timeFactor, is used to calculate the dynamic decay of the relationship strength based on the half-life. The calculation method is as follows: timeFactor = Date Difference / Half-life, where: Date Difference is the absolute value of the date difference between the current time and the last matching time in the matching evidence, and the half-life is usually set to 30 days.

8. The hotel matching method based on multidimensional relational reasoning according to claim 5, characterized in that, The consistency factor is used to determine the consistency between the current match and the previous match result, and is calculated as follows: consistencyFactor = successRate × (1 - Variance); Where successRate is the historical success rate and variance is the variance; Among them, the historical success rate is the ratio of the number of successful matches of the relationship under the same hotel category in the hotel matching knowledge base to the total number of matches; Variance is the degree of dispersion of the results of multiple matchings within a specific region based on the relationship results of the statistically defined multiple dimensions. It is calculated as follows: Variance = Σ(Match results based on the relationship result each time - Average of the matching results based on the relationship result)² / n.

9. The hotel matching method based on multidimensional relational reasoning according to claim 3, characterized in that, The dynamic adjustment of the weights of each dimension includes the following steps: A basic weight is configured for multiple relational dimensions, and the sum of the basic weights is 1.0; the basic weights include: basic weights for semantic dimensions, basic weights for spatial dimensions, basic weights for business dimensions, and basic weights for historical dimensions; The method for dynamically adjusting the weights of each dimension is as follows: New semantic dimension weight = basic semantic dimension weight × semantic relation strength value. New spatial dimension weight = basic spatial dimension weight × spatial relationship strength value. New weight for the business dimension = basic weight for the business dimension × business relationship strength value. New weight for historical dimension = base weight for historical dimension × historical relationship strength value; The adjusted weights of each dimension are normalized to generate semantic dynamic weights, spatial dynamic weights, business matching dynamic weights, and historical dynamic weights.

10. The hotel matching method based on multidimensional relational reasoning according to claim 3, characterized in that, The method for calculating the overall confidence level is as follows: Overall confidence level = Semantic weighted confidence level + Spatial weighted confidence level + Commercial weighted confidence level + Historical weighted confidence level Wherein, semantic weighted confidence = semantic dynamic weight × semantic relationship similarity, Spatial weighted confidence score = Spatial dynamic weight × (1 - straight-line distance / distance standard), Business-weighted confidence score = Business dynamic weight × Business relationship value Historical weighted confidence score = Historical matching weight × (Semantic relationship historical matching success rate + Spatial relationship historical matching success rate + Business relationship historical matching success rate) / 3.

Images