Name matching system and method based on integrated AI algorithm

Abstract

The present invention belongs to the field of intelligent name matching, and in particular relates to a name matching system and method based on an integrated AI algorithm. The system can extract name information to be matched in an input text, and judge the probability that a name is an alias or a wrong name through a discriminant model and an alias mapping library, and analyze the name matching scene features at the same time, and call one or more matching unit combinations of a fuzzy integrated matching unit, a speech coding matching unit, a machine learning model matching unit and a large language model matching unit according to the scene features and the name information to be matched for analysis, so as to obtain the matching confidence and result value of each matching unit; finally, an integrated matching coding feature is constructed through an integrated coding unit, and a final name matching result is output through an output unit; the present invention significantly improves the accuracy and robustness of name matching through the integrated application of a multi-level matching algorithm, and is suitable for a wide range of language environments and complex data scenarios.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent name matching, and in particular relates to a name matching system and method based on an integrated AI algorithm. Background Art

[0002] In different cultural and language environments, there are significant differences in the writing habits, pronunciation rules and cultural backgrounds of names, which makes the automatic matching of names extremely complicated. Traditional matching methods mainly rely on precise string matching or rule-based fuzzy matching, but they are often powerless when facing cross-language name matching; in addition, the existing technology has the problems of large limitations in precise matching, rigid rule-based fuzzy matching rules and difficulty in covering all variants, additional errors introduced during language conversion, and neglect of cultural differences and context understanding; in addition, the existing technical solutions are often limited to specific fields and are limited to character matching. However, if the names are matched only from the perspective of string similarity without considering semantics, it is difficult to regard them as the same person.

[0003] For example, a Chinese patent with authorization announcement number CN104008123B discloses a method for matching Chinese names, which converts Chinese names into Chinese pinyin and matches them with Chinese names spelled in Roman letters, and then applies native script comparison to obtain the final matching score. However, this method mainly relies on the conversion accuracy of Chinese pinyin, and has limitations when dealing with complex name variants and cross-language matching, especially when dealing with non-standard spellings or dialect pronunciations, it is easy to produce mismatches; in addition, a Chinese patent with authorization announcement number CN116010562B discloses a name matching method, device, equipment and medium based on multiple data sources, which improves the matching success rate through speech recognition and multi-dimensional most similar matching algorithm. However, this method still faces challenges in practical applications, such as the accuracy of speech recognition is affected by accents and environmental noise, and the multi-dimensional most similar matching algorithm may increase the computational complexity when processing a large number of data sources, and the performance is unstable when processing unstructured or semi-structured data. Although these two methods have solved the name matching problem in specific scenarios to a certain extent, they still have problems such as low matching accuracy, high processing complexity and insufficient adaptability in cross-language environments and complex application scenarios.

[0004] The above prior art has the following problems: the prior art solutions are often limited to specific fields or scenarios, and are all limited to character matching. However, if the semantics of the name is not considered and only the similarity matching is performed from the perspective of character strings, it is difficult to regard them as the same person, especially in many fields where pen names or stage names are used. For this reason, the present invention provides a name matching system and method based on an integrated AI algorithm. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes a name matching system and method based on an integrated AI algorithm. The system is provided with a name discrimination unit, a name receiving unit and a scene analysis unit, which can extract the name information to be matched in the input text or picture, and discriminate the probability that the name is an alias or a wrong name through a discrimination model and an alias mapping library, and analyze the name matching scene features at the same time, and call one or more matching unit combinations of a fuzzy integrated matching unit, a speech coding matching unit, a machine learning model matching unit and a large language model matching unit according to the scene features and the name information to be matched for analysis, so as to obtain the matching confidence and result value of each matching unit; finally, an integrated matching coding feature is constructed through an integrated coding unit, and the final name matching result is output through an output unit; the present invention significantly improves the accuracy and robustness of name matching through the integrated application of a multi-level matching algorithm, and is suitable for a wide range of language environments and complex data scenarios.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The name matching system based on the integrated AI algorithm includes: a receiving module, an integrated processing module and an output module; the receiving module includes a name recognition unit, a name receiving unit and a scene analysis unit;

[0008] The name receiving unit is used to extract the text information of the name to be matched from the input text; the name distinguishing unit is used to distinguish the probability of the received name being an alias or a wrong name based on the text information of the input name to be matched, through the configured distinguishing model and alias mapping library, and construct the name input to be matched by using the text information of the name to be matched and the distinguishing probability; the scene analysis unit is used to analyze the text information of the input name to be matched, extract the name matching scene features, and use the name matching scene features and the distinguishing probability obtained by the name distinguishing unit to call the matching algorithm integrated in the integrated processing module through the reinforcement scheduling model to perform secondary matching on the constructed name input to be matched to obtain a matching result;

[0009] The integrated processing module includes a fuzzy integrated matching unit, a speech coding matching unit, a machine learning model matching unit and a large language model matching unit;

[0010] The name pair to be matched obtained by the name receiving unit is analyzed by acquiring the name matching scene features and calling one or a combination of n matching units among the fuzzy integrated matching unit, the speech coding matching unit, the machine learning model matching unit and the large language model matching unit, so as to obtain the matching confidence and matching result value corresponding to each matching unit;

[0011] The output module includes an integrated encoding unit and an output unit;

[0012] The matching confidence and matching result values ​​corresponding to each unit called in the integrated processing module are input into the integrated coding unit to construct an integrated matching coding feature, and the name matching result is obtained according to the integrated matching coding feature through the output unit.

[0013] Specifically, the specific steps of the name identification unit for identification include:

[0014] A1. Obtaining the name entity string to be matched and the corresponding contextual relationship information through the name receiving unit;

[0015] A2. Perform alias matching based on the configured alias mapping library and the name entity string to be matched by using a fuzzy matching algorithm. If a unique alias is matched, the current name entity to be matched is an alias. According to the current name entity to be matched and the alias-original name mapping relationship constructed in the alias mapping library, obtain the first name to be matched and the corresponding first matching probability;

[0016] A3. If more than one alias is matched, a one-to-many mapping relationship is established between the name entity to be matched and all the matched alias-original name mapping relationships, and a second list of names to be matched and corresponding matching probabilities are obtained;

[0017] A4. If the number of matched aliases is 0, the name entity to be matched and the corresponding contextual relationship information are input into the fuzzy matching algorithm to match the original name in the alias mapping library, and an initial original name matching list and the probability that each matched original name is a wrong name are obtained.

[0018] Specifically, the step of extracting name matching scene features by the scene analysis unit includes:

[0019] B1. Based on the name entity string to be matched obtained in A1, the discrimination algorithm is used to determine whether the name entity string to be matched is a Chinese string name or an English string name. If it is an English string name, the surname string in the string is used to determine whether the corresponding English string name is an English alias corresponding to a Chinese person or a foreigner's name, and the judgment result is marked on the corresponding name entity string to be matched. If it is a Chinese string, no marking is performed;

[0020] B2. Determine whether the received text information of the name to be matched contains context information according to the text information of the name to be matched. If it contains, repeat A1-A4 and B1 processes to perform matching judgment to obtain the initial matching input of the first scene name containing the context information.

[0021] B3, if not, repeat A1-A3 and B1 process to perform initial matching, and construct the relationship mapping information obtained by matching A2 and A3 with the name to be matched to obtain the initial matching input of the second scene name that does not contain context information;

[0022] B4, if the initial matching process in B3 does not obtain the relationship mapping information, the initial matching input of the name of the third scene without the context information is used;

[0023] B5. According to the B1 process, the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input are identified and labeled to obtain the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input.

[0024] Specifically, the steps of strengthening the scheduling model construction include:

[0025] C1. According to the fuzzy integrated matching unit, the speech coding matching unit, the machine learning model matching unit and the large language model matching unit integrated in the integrated processing module, the labels of the corresponding matching units are set, and the integrated scheduling code is constructed using the labels set for each matching unit;

[0026] C2. Construct the current input state based on the output results of the name recognition unit and the scene analysis unit and the constructed integrated scheduling code , , , ,in, Enter the name of the first scene after the annotation. Enter the initial matching name for the second scene after annotation. Enter the initial match for the third scene name after annotation. and represents the first name to be matched and the corresponding first matching probability in turn; and represents the ath name to be matched in the second list of names to be matched and the corresponding ath matching probability; and represents the probability that the i-th original name matched in the initial original name matching list and the i-th matched original name are wrong names; and They represent the probability that the kth original name and the kth matching original name obtained in process A4 without contextual information are wrong names; It means that the jth discrimination result is obtained by the B1 process. When it is, it means that the name to be matched is a Chinese string name. When it is, it means that the name to be matched is an English string name. When the name to be matched is the English alias corresponding to the Chinese person under the English string name, When , it means that the name to be matched is a foreigner's name under the English string name.

[0027] Specifically, the steps of strengthening the scheduling model construction also include:

[0028] C3. Set the current execution trigger information according to the current input state, including:

[0029] C31, when the input is the first scene name initial matching input and , triggering the first execution action information, that is, calling the fuzzy integration matching unit and the machine learning model matching unit to perform secondary matching;

[0030] C32, when the input is the first scene name initial matching input and , , , or When , the second execution action information is triggered, that is, the fuzzy integrated matching unit, the machine learning model matching unit and the large language model matching unit are called to perform secondary matching;

[0031] C33, when the input is the first scene name initial matching input and , or , triggers the third execution action information, that is, calling the speech coding matching unit, the machine learning model matching unit and the large language model matching unit for secondary matching.

[0032] Specifically, the steps of strengthening the scheduling model construction also include:

[0033] C34, when the input is the second scene name initial matching input and or , triggering the fourth execution action information, that is, calling the fuzzy integration matching unit and the machine learning model matching unit to perform secondary matching;

[0034] C35, when the input is the second scene name initial matching input and or When the fifth execution action information is triggered, the fuzzy integration matching unit, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching;

[0035] C36, when the input is the second scene name initial matching input and or When the sixth execution action information is triggered, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching;

[0036] C37, when the input is the third scene name initial matching input and When the seventh execution action information is triggered, the fuzzy integration matching unit, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching.

[0037] Specifically, the alias includes abbreviations, spelling variations, pen names, stage names or pseudonyms; the steps of constructing the alias mapping library include:

[0038] D1. Obtain basic information of a person's name and corresponding alias information and perform preprocessing;

[0039] D2. Input the preprocessed information into the entity-relationship joint extraction model to obtain the original name of each entity, entity alias and corresponding relationship information;

[0040] D3. Input each entity original name, entity alias and corresponding relationship information into the graph database, with the entity original name as the first-level node and the entity alias as the second-level node;

[0041] D4. When the entity alias corresponding to the entity original name includes an abbreviation, a spelling variation, a pen name, a stage name, or a pseudonym, the pen name, the stage name, or the pseudonym is stored in an independent child node under the secondary node, and the independent child node is connected to the primary node, and the connection relationship is represented by 1;

[0042] D5. When the entity alias corresponding to the entity original name does not contain one or more of the pen name, stage name or alias, the corresponding independent child nodes that are not contained are filled with 0, and the corresponding connection relationship is not marked.

[0043] The name matching method based on the integrated AI algorithm includes the following steps:

[0044] S1. Obtain text information of the name to be matched, obtain the probability that the received name is an alias or a wrong name through the configured discriminant model and alias mapping library, and construct the name input to be matched by using the text information and probability information of the name to be matched;

[0045] S2. Obtain name matching scenario features based on the acquired text information of the name to be matched, and obtain matching confidence and matching result values ​​of the corresponding units by scheduling different matching units integrated in the integrated processing module through the configured enhanced scheduling model based on the name matching scenario features and the acquired discrimination probability;

[0046] S3. An integrated matching coding feature is constructed according to the matching confidence and matching result value of the corresponding unit, and the integrated matching coding feature is input into a support vector machine to obtain a final name matching result.

[0047] A computer-readable storage medium stores computer instructions, which, when executed, execute a name matching method based on an integrated AI algorithm.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] In view of the deficiencies in the prior art, the present invention uses a configured discrimination model and an alias mapping library to discriminate the probability that a name is an alias or a wrong name, and uses context information to construct a more accurate name input to be matched. At the same time, the input information is analyzed to obtain input scene type information, and the name input to be matched is combined with the scene information. Different matching algorithm combinations are scheduled by an enhanced scheduling model to perform name matching. This process can not only significantly improve the accuracy and reliability of matching, effectively avoid the problems of misjudgment and missed judgment caused by simple character matching, but also reduce the waste of memory due to the simultaneous calling of all algorithms, and improve the recognition speed. In addition, through different scenarios and different matching algorithm combinations, the present invention can be applied to more matching scenarios, and the applicability of the present invention is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a module diagram of a name matching system based on an integrated AI algorithm according to Embodiment 1 of the present invention;

[0051] Figure 2 This is a flow chart of the matching process of the fuzzy integrated matching algorithm according to Embodiment 1 of the present invention;

[0052] Figure 3 This is a flow chart of the matching process of the speech coding matching algorithm in Example 1 of the present invention;

[0053] Figure 4 This is a flowchart of the corresponding unit of the name matching system based on the integrated AI algorithm in Example 2 of the present invention;

[0054] Figure 5 This is a flow chart of a name matching method based on an integrated AI algorithm according to Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0055] Example 1

[0056] See also Figure 1 , an embodiment provided by the present invention: a name matching system based on an integrated AI algorithm, comprising: a receiving module, an integrated processing module and an output module;

[0057] The receiving module is used to receive names to be matched in the same language or cross-language scenarios; the receiving module includes a scenario analysis unit, a name receiving unit and a name identification unit;

[0058] The name receiving unit is used to extract text information of the name to be matched from the input text;

[0059] The name discrimination unit is used to discriminate the probability of the received name being an alias or a wrong name according to the text information of the input name to be matched, through the configured discrimination model and alias mapping library, and construct the input name to be matched by using the text information of the name to be matched and the discrimination probability;

[0060] Furthermore, in this embodiment, the specific steps of the name identification unit for identification include:

[0061] A1. Obtaining the name entity string to be matched and the corresponding contextual relationship information through the name receiving unit;

[0062] A2. Perform alias matching based on the configured alias mapping library and the name entity string to be matched by using a fuzzy matching algorithm. If a unique alias is matched, the current name entity to be matched is an alias. According to the current name entity to be matched and the alias-original name mapping relationship constructed in the alias mapping library, obtain the first name to be matched and the corresponding first matching probability;

[0063] Furthermore, in this embodiment, the alias includes abbreviations, spelling variations, pen names, stage names or pseudonyms; the steps of constructing the alias mapping library include:

[0064] D1. Obtain basic information of a person's name and corresponding alias information and perform preprocessing;

[0065] D2. Input the preprocessed information into the entity-relationship joint extraction model to obtain the original name of each entity, entity alias and corresponding relationship information;

[0066] D3. Input each entity original name, entity alias and corresponding relationship information into the graph database, with the entity original name as the first-level node and the entity alias as the second-level node;

[0067] D4. When the entity alias corresponding to the entity original name includes an abbreviation, a spelling variation, a pen name, a stage name, or a pseudonym, the pen name, the stage name, or the pseudonym is stored in an independent child node under the secondary node, and the independent child node is connected to the primary node, and the connection relationship is represented by 1;

[0068] D5. When the entity alias corresponding to the entity original name does not contain one or more of the pen name, stage name or alias, the corresponding independent child nodes that are not contained are filled with 0, and the corresponding connection relationship is not marked.

[0069] A3. If more than one alias is matched, a one-to-many mapping relationship is established between the name entity to be matched and all the matched alias-original name mapping relationships, and a second list of names to be matched and corresponding matching probabilities are obtained;

[0070] A4. If the number of matched aliases is 0, the name entity to be matched and the corresponding contextual relationship information are input into the fuzzy matching algorithm to match the original name in the alias mapping library, and an initial original name matching list and the probability that each matched original name is a wrong name are obtained.

[0071] The scene analysis unit is used to analyze the text information of the input name to be matched, extract the name matching scene features, and use the name matching scene features and the discrimination probability obtained by the name discrimination unit to call different matching algorithms integrated in the integrated processing module through the reinforcement scheduling model to perform secondary matching on the constructed name input to be matched to obtain a matching result;

[0072] Furthermore, in this embodiment, the step of extracting name matching scene features by the scene analysis unit includes:

[0073] B1. Based on the name entity string to be matched obtained in A1, the discrimination algorithm is used to determine whether the name entity string to be matched is a Chinese string name or an English string name. If it is an English string name, the surname string in the string is used to determine whether the corresponding English string name is an English alias corresponding to a Chinese person or a foreigner's name, and the judgment result is marked on the corresponding name entity string to be matched. If it is a Chinese string, no marking is performed;

[0074] B2. Determine whether the received text information of the name to be matched contains context information according to the text information of the name to be matched. If it contains, repeat A1-A4 and B1 processes to perform matching judgment to obtain the initial matching input of the first scene name containing the context information.

[0075] B3, if not, repeat A1-A3 and B1 process to perform initial matching, and construct the relationship mapping information obtained by matching A2 and A3 with the name to be matched to obtain the initial matching input of the second scene name that does not contain context information;

[0076] B4, if the initial matching process in B3 does not obtain the relationship mapping information, the initial matching input of the name of the third scene without the context information is used;

[0077] B5. According to the B1 process, the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input are identified and labeled to obtain the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input.

[0078] Furthermore, in this embodiment, the step of strengthening the scheduling model construction includes:

[0079] C1. According to the fuzzy integrated matching unit, speech coding matching unit, machine learning model matching unit and large language model matching unit integrated in the integrated processing module, the corresponding matching unit labels are set, and the integrated scheduling coding is constructed using the labels set for each matching unit. , where the fuzzy ensemble matching unit is numbered 0, the speech coding matching unit is numbered 1, the machine learning model matching unit is numbered 2, and the large language model matching unit is numbered 3;

[0080] C2. Construct the current input state based on the output results of the name recognition unit and the scene analysis unit and the constructed integrated scheduling code , , , ,in, Enter the name of the first scene after the annotation. Enter the initial matching name for the second scene after annotation. Enter the initial match for the third scene name after annotation. and represents the first name to be matched and the corresponding first matching probability in turn; and represents the ath name to be matched in the second list of names to be matched and the corresponding ath matching probability; and represents the probability that the i-th original name matched in the initial original name matching list and the i-th matched original name are wrong names; and They represent the probability that the kth original name and the kth matching original name obtained in process A4 without contextual information are wrong names; It means that the jth discrimination result is obtained by the B1 process. When it is, it means that the name to be matched is a Chinese string name. When it is, it means that the name to be matched is an English string name. When the name to be matched is the English alias corresponding to the Chinese person under the English string name, When , it means the name to be matched is a foreigner's name under the English string name;

[0081] C3. Set the current execution trigger information according to the current input state, including:

[0082] C31, when the input is the first scene name initial matching input and When , the first execution action information is triggered, that is, the fuzzy integrated matching unit 0 and the machine learning model matching unit 2 are called to perform secondary matching;

[0083] Furthermore, the corresponding references to the fuzzy integrated matching unit 0 and the machine learning model matching unit 2 in this step are because in this scenario the corresponding references only involve the matching of Chinese names and the input information includes the context text information and the relationship information obtained by matching the alias library. Therefore, the fuzzy integrated matching unit 0 and the machine learning model matching unit 2 are called to match and classify the annotated Chinese names.

[0084] C32, when the input is the first scene name initial matching input and , , , or When , the second execution action information is triggered, that is, the fuzzy integrated matching unit 0, the machine learning model matching unit 2 and the large language model matching unit 3 are called to perform secondary matching;

[0085] Furthermore, in this step, the corresponding references to fuzzy integrated matching unit 0, machine learning model matching unit 2 and large language model matching unit 3 are because in this scenario, the corresponding matches all involve Chinese names. According to the characteristics of Chinese matching and the advantage of fuzzy integrated matching algorithm in Chinese matching, fuzzy integrated matching unit 0 is called. However, in this scenario, the matched English matching pseudonyms or stage names and wrong names are included, and the corresponding aliases may not belong to the same person. Therefore, the machine learning model matching unit is called for classification prediction. In addition, in this scenario, feature information of contextual relationship information related to name matching is included, and the first two matching algorithms have limited text processing capabilities. Therefore, the large language model matching unit is called to process the contextual information to further improve the matching accuracy.

[0086] C33, when the input is the first scene name initial matching input and , or When the third execution action information is triggered, the speech coding matching unit 1, the machine learning model matching unit 2 and the large language model matching unit 3 are called to perform secondary matching;

[0087] The reason why the speech coding matching unit is called in this scenario but the fuzzy integrated matching unit is not is that the name matching in this process is all marked in the English name scenario. English names are generally in Latin letters, and the speech coding matching unit is good at Latin letters. Therefore, the speech coding matching unit is called in this process.

[0088] C34, when the input is the second scene name initial matching input and or When , the fourth execution action information is triggered, that is, the fuzzy integrated matching unit 0 and the machine learning model matching unit 2 are called to perform secondary matching;

[0089] C35, when the input is the second scene name initial matching input and or When the fifth execution action information is triggered, the fuzzy integrated matching unit 0, the speech coding matching unit 1 and the machine learning model matching unit 2 are called to perform secondary matching;

[0090] C36, when the input is the second scene name initial matching input and or When the sixth execution action information is triggered, that is, the speech coding matching unit 1 and the machine learning model matching unit 2 are called to perform secondary matching;

[0091] C37, when the input is the third scene name initial matching input and When , the seventh execution action information is triggered, that is, the fuzzy integrated matching unit 0, the speech coding matching unit 1 and the machine learning model matching unit 2 are called to perform secondary matching;

[0092] The large language model matching unit is not called in the C34-C37 process because the second and third scenarios do not contain contextual text information, so a text processing model is not required. Name matching belongs to phrase matching, and the text processing capabilities of the remaining three units can complete the corresponding text processing. This can further reduce the resource load of the model operation and improve the matching speed.

[0093] C4. Execute the trigger information according to the current time, and build the current time execution action ,in, Indicates the action that triggers the first execution action information to be executed. Indicates the action that triggers the second execution action information to be executed. Indicates the action that triggers the execution of the corresponding action of the third execution action information. Indicates the action executed by triggering the fourth execution action information. Indicates the action that triggers the fifth execution action information to be executed. Indicates the action that triggers the sixth execution action information to be executed. Indicates the action executed corresponding to the triggering of the seventh execution action information; C5, constructing an enhanced scheduling model based on the SAC algorithm, inputting the constructed current moment input state, current moment execution trigger information and current moment execution action into the enhanced scheduling model for training, obtaining the trained enhanced scheduling model, and building it into the scene analysis unit, and scheduling the corresponding matching algorithm combination according to the results obtained by the name discrimination unit and the scene analysis unit for name matching processing.

[0094] This process achieves accurate reception and processing of names to be matched in the same language or cross-language scenarios by constructing a receiving module including a scene analysis unit, a name receiving unit and a name discrimination unit. The name discrimination unit uses the discrimination model and the alias mapping library to effectively discriminate the probability of aliases or wrong names, thereby improving the accuracy of name matching; the scene analysis unit, based on the characteristics of the name matching scenario, realizes secondary matching of the input of the name to be matched by intelligently scheduling different matching algorithms in the integrated processing module through the enhanced scheduling model, further optimizing the matching results; in addition, the enhanced scheduling model can match different name matching scenarios by setting trigger information in different scenarios. This intelligent name matching processing method not only improves processing efficiency, but also greatly enhances the accuracy and generalization of matching results, and can simultaneously process the corresponding name matching processes in different types of scenarios.

[0095] The integrated processing module is used to extract and match the obtained names to be matched by using integrated multi-type algorithms to obtain matching results under different types of algorithms;

[0096] The integrated processing module includes a fuzzy integrated matching unit, a speech coding matching unit, a machine learning model matching unit and a large language model matching unit;

[0097] The fuzzy integrated matching unit is used to calculate the fuzzy matching confidence level through the integrated fuzzy matching algorithm according to the obtained name text information to be matched, and obtain the fuzzy matching result value according to the set fuzzy matching threshold;

[0098] There are many fuzzy matching algorithms in the prior art. Each fuzzy matching technology can only evaluate the similarity of strings from one perspective, and each fuzzy matching technology has its own focus and different performances in different scenarios. For example, some focus on the matching degree of the entire string, while others focus on the maximum matching of some substrings; some may emphasize the order of characters, while others weaken the influence of the order of characters; therefore, in this embodiment, the fuzzy integrated matching unit is configured with a fuzzy integrated matching algorithm integrated by multiple fuzzy matching algorithms, please refer to Figure 2 ,The steps of name matching by fuzzy integrated matching algorithm include:

[0099] Get the text information of name 1 and name 2 to be matched;

[0100] The obtained text information of name 1 and name 2 to be matched is input into fuzzy matching algorithm 1 to fuzzy matching algorithm m for matching processing, and the corresponding matching confidence is output at the same time ;

[0101] According to the obtained matching confidence , the total matching confidence is obtained through the weighted algorithm , specifically:

[0102]

[0103] Represents the number of different fuzzy matching algorithms integrated; Represents the weight of the i-th fuzzy matching algorithm and satisfies ; is the standardized confidence level of the ith fuzzy match, satisfying ;

[0104] Set the fuzzy match threshold , based on the total matching confidence and , get the final fuzzy matching result value ;

[0105] ;

[0106] when When the matching condition is not met, the matching result is 0. The matching condition is met, and the matching result is 1.

[0107] The speech coding matching unit is used to convert the original or converted Latin letter string format name to be matched into a sound-shape code through the Metaphone algorithm, and calculate the Levenshtein distance between the sound-shape codes to be matched, and use the calculated distance and the set threshold to obtain the code matching confidence and the sound-shape code matching result value.

[0108] For further information, see Figure 3 In this embodiment, the steps of performing name matching by the voice coding matching unit include:

[0109] Obtain the text information of name 1 and name 2 to be matched, and use a discrimination algorithm to determine whether name 1 and name 2 to be matched are Latin names. If not, convert name 1 or name 2 to be matched into Latin names to obtain lat_name 1 and lat_name 2 in Latin state;

[0110] Input the obtained lat_name1 and lat_name2 in the Latin alphabet state into the Metaphone coding layer to obtain the phonetic-graphic coding 1 and the phonetic-graphic coding 2;

[0111] Using the calculated sound-shape coding 1 and sound-shape coding 2, the coding matching confidence is calculated through the Levenshtein formula. ;

[0112] Set the sound shape matching threshold , the confidence of the encoding match is calculated and , by judging the conditions, the final sound-shape coding matching result value is obtained ; The specific judgment conditions are:

[0113] ;

[0114] Among them, when When the matching condition is not met, the matching result is 0. The matching condition is met, and the matching result is 1 at this time; further, 0 here means no match, 1 means match, and the judgment conditions 0 and 1 in the corresponding units below also mean the same thing.

[0115] The machine learning model matching unit is used to encode the character string features of the obtained name text information to be matched, and use the convolutional attention model to obtain the classification prediction confidence that the names to be matched belong to the same person, and use the built-in classification matching threshold to obtain the classification prediction matching result value;

[0116] Furthermore, in this embodiment, a classification prediction algorithm is configured in the machine learning model matching unit to predict the difference in the name vector values ​​and determine whether they belong to the same person. The steps include:

[0117] Get the text information of name 1 and name 2 to be matched, and use Word Embedding to vectorize the input text information of name 1 and name 2 to be matched to obtain the vectorized matching name and name ;

[0118] The vectorized matching names and name , input into the difference discrimination model constructed by random forest, obtain the difference prediction result, and use the difference prediction result as the classification prediction confidence ;

[0119] Setting the classification match threshold in the difference discriminant model , according to the classification prediction confidence and , through the discriminant conditions, obtain the classification prediction matching result value ; The judgment conditions are:

[0120] ;

[0121] Among them, when When the matching condition is not met, the matching result is 0. The matching condition is met, and the matching result is 1.

[0122] The large language model matching unit is used to semantically encode the input name text information to be matched, obtain the model matching confidence through the configured language model, and obtain the classification matching result value through the built-in model matching threshold;

[0123] Furthermore, the matching algorithms built into the fuzzy integration matching unit, the speech coding matching unit, and the machine learning model matching unit focus more on the comparison of string similarity, and have weak support for semantics. They do not provide sufficient support for words with large spelling differences but the same semantics. If semantics are not considered and only similarity matching is performed from the perspective of strings, it is difficult to regard them as the same person. Therefore, in this embodiment, a Bert model based on pre-training in Chinese and English is matched in the large language model matching unit to perform semantic recognition in the process of name matching. The matching steps include:

[0124] Get the text information of name 1 and name 2 to be matched;

[0125] Input the text information of name 1 and name 2 to be matched into the configured Chinese and English pre-trained Bert model to obtain the model matching confidence ;

[0126] Set model matching threshold , according to the model matching confidence and , through the judgment conditions, obtain the classification matching result value ; The judgment conditions are:

[0127]

[0128] Among them, when When the matching condition is not met, the matching result is 0. The matching condition is met, and the matching result is 1.

[0129] The output module is used to construct an integrated matching vector according to the different matching results obtained, and obtain the final name matching result through a support vector machine; the output module includes an integrated encoding unit and an output unit;

[0130] The integrated coding unit is used to construct an integrated matching coding feature according to different matching confidences and matching result values ​​output by the integrated processing module;

[0131] Furthermore, in this embodiment, the specific steps of constructing the integrated matching coding feature include:

[0132] According to the results obtained by the scene analysis unit and the name receiving unit, the matching confidence and matching result value output by the corresponding matching unit are scheduled through the reinforcement scheduling model to construct an integrated matching coding feature;

[0133] Assuming that the units scheduled by the enhanced scheduling model are fuzzy integrated matching units, machine learning model matching units, and large language model matching units, the corresponding integrated matching encoding features are ;

[0134] Furthermore, in order to keep the integrated matching coding features input to the output unit consistent, the fixed length of the integrated matching coding features is set to 9. When any of the fuzzy integrated matching units, speech coding matching units, machine learning model matching units and large language model matching units is not scheduled, the output feature vector of the unscheduled matching unit in the integrated matching coding features is represented by 0, so that the dimensions of the integrated matching coding features corresponding to all inputs to the output units remain consistent in real time.

[0135] The output unit is used to obtain the name matching result through the support vector machine according to the integrated matching coding features constructed by the different matching algorithms called by the scene analysis unit.

[0136] Example 2

[0137] See also Figure 4 Another embodiment provided by the present invention is a corresponding unit workflow of a name matching system based on an integrated AI algorithm, including:

[0138] First, the text information of the name to be matched is received by the name receiving unit, and the obtained text information of the name to be matched is input into the name distinguishing unit, the received name is distinguished and the probability that the received name is an alias or a wrong name is obtained, and the text information of the name to be matched and the probability information are used to construct the input of the name to be matched;

[0139] Second, the text information of the input name to be matched is analyzed by the scene analysis unit to extract the name matching scene features, and the name matching scene features and the discrimination probability obtained by the name discrimination unit are used to call one or more unit combinations of the fuzzy integrated matching unit, the speech coding matching unit, the machine learning model matching unit and the large language model matching unit to perform secondary matching on the constructed name input to be matched to obtain the matching confidence and matching result value of the corresponding unit; the matching confidence and matching result value of the corresponding unit are input into the integrated coding unit to construct the integrated matching coding feature, and the constructed integrated matching coding feature is input into the output unit, and the final name matching result is obtained through the support vector machine.

[0140] Example 3

[0141] See also Figure 5 Another embodiment provided by the present invention is a name matching method based on an integrated AI algorithm, comprising:

[0142] S1. Obtain text information of the name to be matched, obtain the probability that the received name is an alias or a wrong name through the configured discriminant model and alias mapping library, and construct the name input to be matched by using the text information and probability information of the name to be matched;

[0143] S2. Obtain name matching scenario features based on the acquired text information of the name to be matched, and obtain matching confidence and matching result values ​​of the corresponding units by scheduling different matching units integrated in the integrated processing module through the configured enhanced scheduling model based on the name matching scenario features and the acquired discrimination probability;

[0144] S3. An integrated matching coding feature is constructed according to the matching confidence and matching result value of the corresponding unit, and the integrated matching coding feature is input into a support vector machine to obtain a final name matching result.

[0145] Example 4

[0146] A computer-readable storage medium stores computer instructions, which, when executed, execute a name matching method based on an integrated AI algorithm.

[0147] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a name matching method based on an integrated AI algorithm when executing the computer program.

[0148] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

[0149] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device for processing personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A name matching system based on integrated AI algorithm, characterized in that: include: A receiving module, an integrated processing module and an output module; the receiving module includes a name identification unit, a name receiving unit and a scene analysis unit; The name receiving unit is used to extract text information of the name to be matched from the input text; the name distinguishing unit is used to distinguish the probability of the received name being an alias or a wrong name based on the text information of the input name to be matched, through the configured distinguishing model and alias mapping library, and construct the name input to be matched by using the text information of the name to be matched and the distinguishing probability; the scene analysis unit is used to analyze the text information of the input name to be matched, extract the name matching scene features, and use the name matching scene features and the distinguishing probability obtained by the name distinguishing unit to call the matching algorithm integrated in the integrated processing module through the reinforcement scheduling model to perform secondary matching on the constructed name input to be matched to obtain a matching result; The integrated processing module includes a fuzzy integrated matching unit, a speech coding matching unit, a machine learning model matching unit and a large language model matching unit; The name pair to be matched obtained by the name receiving unit is analyzed by acquiring the name matching scene features and calling one or a combination of n matching units among the fuzzy integrated matching unit, the speech coding matching unit, the machine learning model matching unit and the large language model matching unit to obtain the matching confidence and matching result value corresponding to each matching unit; The output module includes an integrated encoding unit and an output unit; The matching confidence and matching result values ​​corresponding to each unit called in the integrated processing module are input into the integrated coding unit to construct an integrated matching coding feature, and the name matching result is obtained according to the integrated matching coding feature through the output unit.

2. The name matching system based on integrated AI algorithm as claimed in claim 1, characterized in that: The specific steps of the name identification unit for identification include: A1. Obtaining the name entity string to be matched and the corresponding contextual relationship information through the name receiving unit; A2. Perform alias matching based on the configured alias mapping library and the name entity string to be matched by using a fuzzy matching algorithm. If a unique alias is matched, the current name entity to be matched is an alias. According to the current name entity to be matched and the alias-original name mapping relationship constructed in the alias mapping library, obtain the first name to be matched and the corresponding first matching probability; A3. If more than one alias is matched, a one-to-many mapping relationship is established between the name entity to be matched and all the matched alias-original name mapping relationships, and a second list of names to be matched and corresponding matching probabilities are obtained; A4. If the number of matched aliases is 0, the name entity to be matched and the corresponding contextual relationship information are input into the fuzzy matching algorithm to match the original name in the alias mapping library, and an initial original name matching list and the probability that each matched original name is a wrong name are obtained.

3. The name matching system based on integrated AI algorithm as claimed in claim 2, characterized in that: The step of extracting name matching scene features by the scene analysis unit comprises: B1. According to the name entity string to be matched obtained in A1, through the discrimination algorithm, it is determined that the name entity string to be matched is a Chinese string name or an English string name. If it is an English string name, the corresponding English string name is determined to be an English alias corresponding to a Chinese person or a foreigner's name through the surname string in the string, and the judgment result is marked on the corresponding name entity string to be matched. If it is a Chinese string, no marking is performed; B2. Determine whether the received text information of the name to be matched contains context information according to the text information of the name to be matched. If it contains, repeat A1-A4 and B1 processes to perform matching judgment to obtain the initial matching input of the first scene name containing the context information. B3, if not, repeat A1-A3 and B1 process to perform initial matching, and construct the relationship mapping information obtained by matching A2 and A3 with the name to be matched to obtain the initial matching input of the second scene name that does not contain context information; B4, if the initial matching process in B3 does not obtain the relationship mapping information, the initial matching input of the name of the third scene without the context information is used; B5. According to the B1 process, the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input are identified and labeled to obtain the first scene name initial matching input, the second scene name initial matching input and the third scene name initial matching input.

4. The name matching system based on integrated AI algorithm as claimed in claim 3, characterized in that: The steps of building the enhanced scheduling model include: C1. According to the fuzzy integrated matching unit, the speech coding matching unit, the machine learning model matching unit and the large language model matching unit integrated in the integrated processing module, the labels of the corresponding matching units are set, and the integrated scheduling code is constructed using the labels set for each matching unit; C2. Construct the current input state based on the output results of the name recognition unit and the scene analysis unit and the constructed integrated scheduling code , , , ,in, Enter the name of the first scene after the annotation. Enter the initial matching name for the second scene after annotation. Enter the initial match for the third scene name after annotation. and represents the first name to be matched and the corresponding first matching probability in turn; and represents the ath name to be matched in the second list of names to be matched and the corresponding ath matching probability; and represents the probability that the i-th original name matched in the initial original name matching list and the i-th matched original name are wrong names; and They represent the probability that the kth original name and the kth matching original name obtained in process A4 without contextual information are wrong names; It means that the jth discrimination result is obtained by the B1 process. When it is, it means that the name to be matched is a Chinese string name. When it is, it means that the name to be matched is an English string name. When the name to be matched is the English alias corresponding to the Chinese person under the English string name, When , it means that the name to be matched is a foreigner's name under the English string name.

5. The name matching system based on integrated AI algorithm as claimed in claim 4, characterized in that: The step of strengthening the scheduling model construction also includes: C3. Setting the current moment execution trigger information according to the current moment input state, including: C31, when the input is the first scene name initial matching input and , triggering the first execution action information, that is, calling the fuzzy integration matching unit and the machine learning model matching unit to perform secondary matching; C32, when the input is the first scene name initial matching input and , , , or When , the second execution action information is triggered, that is, the fuzzy integrated matching unit, the machine learning model matching unit and the large language model matching unit are called to perform secondary matching; C33, when the input is the first scene name initial matching input and , or , triggers the third execution action information, that is, calling the speech coding matching unit, the machine learning model matching unit and the large language model matching unit for secondary matching.

6. The name matching system based on integrated AI algorithm as claimed in claim 5, characterized in that: The step of strengthening the scheduling model construction also includes: C34, when the input is the second scene name initial matching input and or , triggering the fourth execution action information, that is, calling the fuzzy integration matching unit and the machine learning model matching unit to perform secondary matching; C35, when the input is the second scene name initial matching input and or When the fifth execution action information is triggered, the fuzzy integration matching unit, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching; C36, when the input is the second scene name initial matching input and or When the sixth execution action information is triggered, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching; C37, when the input is the third scene name initial matching input and When the seventh execution action information is triggered, the fuzzy integration matching unit, the speech coding matching unit and the machine learning model matching unit are called to perform secondary matching.

7. The name matching system based on integrated AI algorithm as claimed in claim 6, characterized in that: The alias includes an abbreviation, a spelling variation, a pen name, a stage name or a pseudonym; the steps of constructing the alias mapping library include: D1. Obtain basic information of a person's name and corresponding alias information and perform preprocessing; D2. Input the preprocessed information into the entity-relationship joint extraction model to obtain the original name of each entity, entity alias and corresponding relationship information; D3. Input each entity original name, entity alias and corresponding relationship information into the graph database, with the entity original name as the first-level node and the entity alias as the second-level node; D4. When the entity alias corresponding to the entity original name includes an abbreviation, a spelling variation, a pen name, a stage name, or a pseudonym, the pen name, the stage name, or the pseudonym is stored in an independent child node under the secondary node, and the independent child node is connected to the primary node, and the connection relationship is represented by 1; D5. When the entity alias corresponding to the entity original name does not contain one or more of the pen name, stage name or alias, the corresponding independent child nodes that are not contained are filled with 0, and the corresponding connection relationship is not marked.

8. A name matching method based on an integrated AI algorithm, which is implemented based on a name matching system based on an integrated AI algorithm as claimed in any one of claims 1 to 7, characterized in that: include: S1. Obtain text information of the name to be matched, obtain the probability that the received name is an alias or a wrong name through the configured discriminant model and alias mapping library, and construct the name input to be matched by using the text information and probability information of the name to be matched; S2. Obtain name matching scenario features based on the acquired text information of the name to be matched, and obtain matching confidence and matching result values ​​of the corresponding units by scheduling different matching units integrated in the integrated processing module through the configured enhanced scheduling model based on the name matching scenario features and the acquired discrimination probability; S3. An integrated matching coding feature is constructed according to the matching confidence and matching result value of the corresponding unit, and the integrated matching coding feature is input into a support vector machine to obtain a final name matching result.

9. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, the name matching method based on integrated AI algorithm described in claim 8 is executed.

Images