Knowledge conflict processing method and refrigeration equipment

Abstract

The invention provides a knowledge conflict processing method and refrigeration equipment. The method comprises the steps that m pieces of knowledge data to be processed are acquired; respectively calculating the semantic similarity between each item of knowledge data and the rest (m-1) items of knowledge data, and screening the knowledge data which are semantic similar to the knowledge data to construct similar semantic data sets so as to determine m groups of similar semantic data sets; and calculating and selectively combining the knowledge data according to an overlapping rate among the m groups of similar semantic data sets, and determining a target knowledge data set. According to the knowledge conflict processing method provided by the invention, the relevance between the knowledge data can be captured by calculating the semantic similarity, and integration of the knowledge data is facilitated; the knowledge data sets are selectively combined based on the overlapping rate, repeated and redundant knowledge can be avoided, the simplified and non-repeated target knowledge data sets are extracted, and the quality of the knowledge data sets is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a knowledge conflict processing method and a refrigeration device. Background Art

[0002] With the development of artificial intelligence technology, its application has gradually entered all walks of life and even the lives of ordinary consumers. Among them, the application of dialogue systems represented by intelligent question and answer and voice assistants is particularly prominent. Many companies' customer service systems have introduced intelligent question and answer systems to enhance customer service capabilities or reduce labor costs. Since many business departments of companies have accumulated a large number of FAQs (Frequently Asked Questions), FAQ-based text matching algorithms are usually used as the basic means of building intelligent question and answer systems. However, there are also many problems with the FAQ resources, the most prominent of which is the knowledge conflict problem. For example, the question expressions in the FAQ are different, but their semantics are similar, corresponding to multiple answers, and the multiple answers are different from each other, and even inconsistent in some key values.

[0003] At present, the following two methods can be used to solve the above-mentioned knowledge conflict problem: First, try to solve the knowledge conflict problem at the production end of FAQ. That is, when entering new knowledge, use algorithms to compare it with the old knowledge already in the FAQ knowledge base, select several old knowledge that may be similar, and let business personnel manually review whether there is a conflict problem. This method can only be used to solve the problem of incremental knowledge, and is not suitable for the problem of existing knowledge. The second is to abstractly represent the problem, that is, to artificially add an abstract representation form to the problem in FAQ, for example, extract the subject, predicate, etc. and combine them into a sequence according to a certain format, and replace the words in it with standard words. In this way, when different problem expressions expressing the same semantics are used, their abstract representation forms should be the same or at least very similar, so that knowledge conflict problems can be automatically monitored. However, this type of method has high requirements for both manual and machine, which will add a lot of cost to the project itself, and there are certain difficulties in practical application. Summary of the invention

[0004] One of the purposes of the present invention is to provide a method for handling knowledge conflicts so as to solve the problem that the prior art adopts the method of abstracting natural language into a highly structured representation form in solving knowledge conflicts. As a result, no matter whether manual annotation or machine algorithm is used, there are technical problems that the conversion quality cannot be guaranteed and the conversion cost is too high.

[0005] One of the objects of the present invention is to provide a refrigeration device.

[0006] In order to achieve one of the above-mentioned purposes of the invention, the present invention provides a knowledge conflict processing method, including: obtaining m items of knowledge data to be processed; respectively calculating the semantic similarity between each item of knowledge data and the remaining m-1 items of knowledge data, screening knowledge data with semantic similarity to the knowledge data to construct similar semantic data sets, so as to determine m groups of similar semantic data sets; calculating and selectively merging the knowledge data according to the overlap rate between the m groups of similar semantic data sets, so as to determine a target knowledge data set; wherein the target knowledge data set contains n items of knowledge data without overlap; m≥n.

[0007] As a further improvement of an embodiment of the present invention, the knowledge data includes all questions in a preset question and answer knowledge base; the preset question and answer knowledge base includes at least one of several groups of questions raised by users and several corresponding groups of answers.

[0008] As a further improvement of an implementation mode of the present invention, the knowledge data is stock question data.

[0009] As a further improvement of an embodiment of the present invention, the "respectively calculating the semantic similarity between each knowledge data and the remaining m-1 knowledge data" specifically includes: vectorizing each knowledge data to obtain a corresponding knowledge vector; based on a similarity vector retrieval tool, calculating the semantic similarity between each knowledge vector and the remaining m-1 knowledge vectors to obtain m-1 semantic similarity values; wherein the similarity vector retrieval tool includes Faiss similarity search.

[0010] As a further improvement of an embodiment of the present invention, the "screening knowledge data that is semantically similar to the knowledge data to construct a similar semantic data set" specifically includes: based on the m-1 semantic similarity values ​​corresponding to the first knowledge data, judging and determining all remaining knowledge data whose semantic similarity values ​​are greater than a preset similarity threshold, to form a first reference data set; screening the first reference data set to obtain K question data that are ranked top after arranging them in descending order of semantic similarity values, to form the first similar semantic data set; m≥K.

[0011] As a further improvement of an embodiment of the present invention, the "calculating and based on the overlap rate between the m groups of similar semantic data sets" specifically includes: obtaining a first similar semantic data set and a second similar semantic data set; calculating and counting the number of element intersections between the first similar semantic data set and the second similar semantic data set to obtain the amount of overlapping data; comparing the number of elements in the first similar semantic data set and the second similar semantic data set, and counting the number of data sets with a larger number of elements as the merged data amount; calculating the overlap rate corresponding to the first similar semantic data set and the second similar semantic data set based on the overlapping data amount and the merged data amount; wherein the overlap rate is numerically equal to the ratio of the overlapping data amount to the merged data amount.

[0012] As a further improvement of an embodiment of the present invention, the "calculating and selectively merging the knowledge data based on the overlap rate between the m groups of similar semantic data sets to determine the target knowledge data set" specifically includes: respectively calculating the overlap rate between each group of similar semantic data sets and the remaining m-1 groups of similar semantic data sets, screening the similar semantic data sets with the maximum overlap rate to determine the data sets to be merged; wherein, the data sets to be merged include two groups of similar semantic data sets; judging whether the maximum overlap rate is greater than the overlap rate threshold; if so, performing a set union operation on the data sets to be merged, and updating the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets; iteratively calculating and selectively merging the intermediate knowledge data sets according to the overlap rate between each pair of intermediate knowledge data sets until the re-determined maximum overlap rate is not greater than the overlap rate threshold, and determining the target knowledge data set.

[0013] As a further improvement of an embodiment of the present invention, the data set to be merged includes a first similar semantic data set and a second similar semantic data set; the "performing a set union operation on the data set to be merged, and updating the m groups of similar semantic data sets according to the operation results to obtain m-1 groups of intermediate knowledge data sets" specifically includes: merging the data elements of the second similar semantic data set into the first similar semantic data set, updating the first similar semantic data set, and deleting the second similar semantic data set to obtain the m-1 groups of intermediate knowledge data sets.

[0014] As a further improvement of an embodiment of the present invention, the data set to be merged includes a first similar semantic data set and a second similar semantic data set; the "performing a set union operation on the data set to be merged, and updating the m groups of similar semantic data sets according to the operation results to obtain m-1 groups of intermediate knowledge data sets" specifically includes: constructing an empty new similar semantic data set, merging the first similar semantic data set and the second similar semantic data set, storing the merged result to the new similar semantic data set, and deleting the first similar semantic data set and the second similar semantic data set to obtain the m-1 groups of intermediate knowledge data sets.

[0015] As a further improvement of an embodiment of the present invention, after the "calculating and selectively merging the knowledge data based on the overlap rate between the m groups of similar semantic data sets to determine the target knowledge data set", the method also includes: obtaining answer data corresponding to the target knowledge data set; based on the answer data, performing an overlap check on the target knowledge data set to determine whether there are still overlapping knowledge data items in the target knowledge data set under the constraints of the answer data; if so, calculating and merging similar semantic data sets with the largest overlap rate based on the judgment result to update the target knowledge data set.

[0016] To achieve one of the above-mentioned purposes of the invention, the present invention further provides a refrigeration device, comprising: a memory for storing executable instructions; and a processor for implementing any of the steps of the above-mentioned knowledge conflict handling method when running the executable instructions stored in the memory.

[0017] Compared with the prior art, the embodiments of the present invention have at least one of the following beneficial effects:

[0018] The present invention adopts a knowledge conflict processing method, and by calculating semantic similarity to capture the correlation between knowledge data, the knowledge data can be grouped at the semantic level to form multiple groups of similar semantic data sets. It is helpful for the integration of knowledge data, so as to automatically generate multiple groups of knowledge data sets with higher semantic similarity and more aggregation, and provide high-quality data support for subsequent knowledge detection; at the same time, by calculating the overlap rate of similar semantic data sets and selectively merging knowledge data sets, it can avoid the generation of too much repeated and redundant knowledge, extract a streamlined and non-repetitive target knowledge data set, and improve the quality and efficiency of the knowledge data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the steps of a knowledge conflict handling method in one embodiment of the present invention.

[0020] Figure 2 It is a schematic diagram of part of step S2 of the knowledge conflict handling method in one embodiment of the present invention.

[0021] Figure 3 It is a schematic diagram of some steps of the knowledge conflict handling method in the first embodiment of the present invention.

[0022] Figure 4 It is a schematic diagram of some steps of the knowledge conflict handling method in the second embodiment of the present invention.

[0023] Figure 5 It is a schematic diagram of some steps of the knowledge conflict handling method in the third embodiment of the present invention.

[0024] Figure 6 It is a flow chart of a preferred embodiment of a knowledge conflict handling method in one embodiment of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by a person skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0026] It should be noted that the term "comprises" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In addition, the terms "first", "second", etc. are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0027] The reason why the knowledge conflict becomes a problem is that: on the one hand, in terms of user experience, when a user asks the same question, they may get a variety of different answers just because of different statements or different expressions, which can easily confuse the user. On the other hand, it is not conducive to development at the technical level. At present, artificial intelligence has fully turned to deep learning, and deep learning training models can construct positive and negative samples to perform text matching tasks. For example, sample A and sample B are similar (positive samples), and sample A and sample C are not similar (negative samples). If there is a problem of knowledge conflict, similar knowledge data will be classified as negative samples. In this way, the data is not logically smooth, which will also cause confusion in model training, resulting in poor results of the final trained model. Therefore, in the field of natural language processing, solving the problem of knowledge conflict is of great practical significance.

[0028] Based on this, the present invention provides a knowledge conflict processing method, such as Figure 1 As shown, the specific steps include:

[0029] Step S1, obtaining m items of knowledge data to be processed;

[0030] Step S2, respectively calculating the semantic similarity between each knowledge data and the remaining m-1 knowledge data, screening the knowledge data with similar semantics to the knowledge data to construct similar semantic data sets, so as to determine m groups of similar semantic data sets;

[0031] Step S3, calculating and selectively merging the knowledge data according to the overlap rate between the m groups of similar semantic data sets to determine the target knowledge data set.

[0032] In this way, by calculating the semantic similarity to capture the correlation between knowledge data, knowledge data can be grouped at the semantic level, which is helpful for the integration of knowledge data, so as to automatically generate knowledge data set groups with higher semantic similarity and more aggregation, and provide high-quality data support for subsequent knowledge detection; in addition, by calculating the overlap rate of similar semantic data sets and selectively merging knowledge data sets, it is possible to avoid generating too much repeated and redundant knowledge, extract concise and non-repetitive target knowledge data sets, and improve the quality of knowledge data sets.

[0033] The knowledge data may include all questions in a preset question-and-answer knowledge base, and may also include some questions in the preset question-and-answer knowledge base. The preset question-and-answer knowledge base includes at least one of a plurality of groups of questions raised by users and a plurality of corresponding groups of answers.

[0034] In this way, the coverage of knowledge data can be freely adjusted according to needs, with strong flexibility; moreover, the existing knowledge data in the preset question-and-answer knowledge base can be directly called without the need for natural language understanding and knowledge extraction, which simplifies the difficulty of technical implementation.

[0035] In one embodiment of the present invention, the preset question and answer knowledge base includes a FAQ (Frequently Asked Questions) library. Specifically, the FAQ library can be composed of historical questions or real-time questions and corresponding answers generated by users when using the retrieval system or consulting with human customer service; and different expressions are made based on the same question in the initial FAQ to achieve further expansion of the FAQ library, and data collection and data expansion can be further achieved on this basis.

[0036] In addition, in one implementation, each FAQ question in the FAQ library can be regarded as a conceptual entity to express the concept of a type of question; there can be several relationships between the FAQ questions, such as synonym relationships, hierarchical relationships, domain correspondence relationships, etc. Based on this, the originally independent FAQ questions can be integrated into structured knowledge data.

[0037] Preferably, the knowledge data is stock data. In this way, the knowledge data can cover different aspects of various fields, with a large amount of data and wide coverage; directly using stock data can save the cost of obtaining knowledge data, and can also continuously expand and improve the preset question and answer knowledge base, with good scalability and wide coverage.

[0038] In one embodiment, for step S2, the present invention provides a refinement step, first calculating the semantic similarity between each knowledge data and the remaining knowledge data to obtain multiple groups of semantic similarity values; then judging and forming a semantic similarity set that meets the conditions based on each similarity value, thereby realizing grouping and aggregation of multiple knowledge data.

[0039] Specifically, if Figure 2 As shown, in the first embodiment, the part of "respectively calculating the semantic similarity between each knowledge data and the remaining m-1 knowledge data" in step S2 may specifically include the following steps:

[0040] Step S21, vectorizing each knowledge data to obtain a corresponding knowledge vector;

[0041] Step S22, based on the similarity vector retrieval tool, calculate the semantic similarity between each knowledge vector and the remaining m-1 knowledge vectors to obtain m-1 semantic similarity values.

[0042] In this way, by vectorizing the data, it is convenient for the subsequent calculation of semantic similarity and knowledge reasoning; in addition, by using similarity vector retrieval tools, similarity calculation can be performed efficiently on a large scale, with simple operation and high accuracy, which facilitates the subsequent classification and clustering of knowledge data.

[0043] Among them, the similarity vector retrieval tool includes FAISS (Facebook AI Similarity Search) similarity search. The similarity vector retrieval tool is a tool for concentrated and fast search of vectors most similar to the query vector in large-scale vector data. The Faiss similarity search is an open source similarity vector retrieval tool library that can provide efficient similarity search and clustering for dense vectors, support searches of billions of vectors, and index based on clustering results to improve search efficiency. It contains a variety of algorithms for searching vector sets of arbitrary sizes, as well as code support for algorithm evaluation and parameter adjustment, and can also provide external interfaces, wherein the size of the vector set is determined by the RAM memory size. The vectorized representation can be a method of representing text or knowledge in the form of a mathematical vector, that is, the knowledge vector can refer to a vector that uses vectorization to express knowledge data.

[0044] Continue to refer to Figure 2As shown, in the second embodiment, the part of "screening knowledge data with semantic similarity to the knowledge data to construct a similar semantic data set" in step S2 may specifically include the following steps:

[0045] Step S23, judging and determining all remaining knowledge data whose semantic similarity values ​​are greater than a preset similarity threshold value according to the m-1 semantic similarity values ​​corresponding to the first knowledge data, to form a first reference data set;

[0046] Step S24, screening out K items of question data with the highest ranking after arranging them in descending order of semantic similarity values ​​from the first reference data set, to form the first similar semantic data set; m≥K.

[0047] In this way, the top K knowledge data items most semantically relevant to the first knowledge data are screened by using a preset similarity threshold, so that subsequent processing is concentrated on the most similar knowledge, improving work efficiency and balancing quality and quantity.

[0048] It should be noted that the first knowledge data may refer to any one of the m items of knowledge data; in one embodiment of the present invention, the semantic similarity between each item of knowledge data and the other m-1 items of knowledge data is calculated respectively, and the semantic similarities are screened to form several first reference data sets; then, the semantic similarity values ​​corresponding to the first reference data sets are sorted from high to low, and the top K items of knowledge data are selected to group and construct m groups of first similar semantic data sets.

[0049] Preferably, the top 10 knowledge data are selected to construct several groups of the first similar semantic data sets. Of course, the present invention does not exclude other numbers of top-ranked knowledge data, which can be freely adjusted according to the actual application scenario. It can be seen that the adaptive modifications made by those skilled in the art that do not deviate from the concept of the present invention are all included in the scope of the present invention.

[0050] Combination Figure 1 and Figure 3 As shown, in the first embodiment of the present invention, for the part of "calculating and based on the overlap rate between the m groups of similar semantic data sets" in step S3, the knowledge conflict processing method may specifically include the following steps:

[0051] Step S311, obtaining a first similar semantic data set and a second similar semantic data set;

[0052] Step S312, calculating and counting the number of element intersections between the first similar semantic data set and the second similar semantic data set to obtain an amount of overlapping data;

[0053] Step S313, comparing the number of elements in the first similar semantic data set and the second similar semantic data set, and counting the number of data sets with a larger number of elements as the merged data amount;

[0054] Step S314, calculating the overlap ratio corresponding to the first similar semantic data set and the second similar semantic data set based on the overlapping data amount and the merged data amount; wherein the overlapping ratio is numerically equal to the ratio of the overlapping data amount to the merged data amount.

[0055] In this way, by counting the number of intersection elements of two similar semantic data sets, the similarity between the two data sets can be directly quantified; in addition, by selecting the largest data set as the merged data volume, a data set with more comprehensive knowledge coverage can be used; therefore, the overlap rate is calculated based on the intersection data volume and the merged data volume, and the calculation method is simple, the calculation complexity is low, and the accuracy is high.

[0056] The size of the overlap rate is positively correlated with the ratio of similarity of semantically similar data sets. In other words, the larger the overlap rate, the closer the semantics of the two data sets are, and the more similar knowledge data there are, and vice versa. Preferably, the preset threshold of the overlap rate can be set to 70%. Of course, it can be flexibly adjusted according to the actual application scenario, and the present invention does not impose any specific restrictions.

[0057] Furthermore, in order to solve the conflict problem in knowledge data, similar semantic data sets with overlapping rates that meet the conditions are merged, the overlapping knowledge data are deleted, and only one version of the overlapping knowledge data is retained. At the same time, all non-overlapping knowledge data items are retained, so that there is no overlap and redundancy in the knowledge data set, thereby ensuring the quality of the knowledge data.

[0058] Based on this, combined Figure 1 and Figure 4 As shown, in the second embodiment of the present invention, the knowledge conflict handling method may specifically include the following steps:

[0059] Step S321, respectively calculating the overlap rate between each group of similar semantic data sets and the remaining m-1 groups of similar semantic data sets, and selecting the similar semantic data set with the maximum overlap rate to determine the data set to be merged;

[0060] Step S322, determining whether the maximum overlap rate is greater than an overlap rate threshold;

[0061] If yes, jump to step S323, perform a set union operation on the data sets to be merged, and update the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets;

[0062] Step S324, iteratively looping and calculating, and selectively merging the intermediate knowledge data sets according to the overlap rates between each other, until the re-determined maximum overlap rate is not greater than the overlap rate threshold, and determining the target knowledge data set.

[0063] The data set to be merged includes two sets of similar semantic data sets. Thus, by calculating and selectively merging the knowledge data sets according to the overlap rate, it is possible to ensure that the merged knowledge data is highly relevant to the original data sets, improve the knowledge relevance, and expand the coverage of the knowledge data.

[0064] Preferably, the overlap rate threshold can be set to 60%. Of course, the present invention does not exclude other overlap rate threshold value configurations, which can be adaptively adjusted according to actual application scenarios. It can be seen that adaptive modifications made by those skilled in the art that do not deviate from the concept of the present invention are all included in the scope of the present invention.

[0065] Specifically, the data set to be merged can point to two sets of similar semantic data sets with the largest overlap rate. When the overlap rate of the data set to be merged is greater than the overlap rate threshold, the two sets of similar semantic data sets are merged, some overlapping elements in the similar semantic data sets are deleted, and only one version of overlapping knowledge data is retained, and the knowledge data sets are streamlined round by round to improve the efficiency of the next round of overlap rate calculation; at the same time, completely different partial elements are merged into one data set to enrich and expand the coverage of the knowledge data set.

[0066] In one embodiment, the data set to be merged includes a first similar semantic data set and a second similar semantic data set. When the overlap rate of the data sets to be merged is greater than an overlap rate threshold, the second similar semantic data set is merged into the first similar semantic data set, and the second similar semantic data set is deleted. Alternatively, a new similar semantic data set is reconstructed, the merge result of the first similar semantic data set and the second similar semantic data set is stored in the new similar semantic data set, and the first similar semantic data set and the second similar semantic data set are deleted.

[0067] Specifically, in the first embodiment, the part of "performing a set union operation on the data sets to be merged, and updating the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets" in step S323 may specifically include the following steps:

[0068] Step S323': merge the data elements of the second similar semantic data set into the first similar semantic data set, update the first similar semantic data set, and delete the second similar semantic data set to obtain the m-1 groups of intermediate knowledge data sets.

[0069] In this way, there is no need to create a new data set, saving storage space, and directly merging based on the first similar semantic data set can avoid repeated data processing work, shorten the data processing time, and increase the merging speed, thereby improving data processing efficiency.

[0070] In the second embodiment, the part of "performing a set union operation on the data sets to be merged, and updating the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets" in step S323 may specifically include the following steps:

[0071] Step S323", construct an empty new similar semantic dataset, merge the first similar semantic dataset and the second similar semantic dataset, store the merged result in the new similar semantic dataset, and delete the first similar semantic dataset and the second similar semantic dataset to obtain the m-1 groups of intermediate knowledge datasets.

[0072] In this way, without destroying the original first similar semantic data set and the second similar semantic data set, the storage location and storage format of the new data set can be freely selected, and the data can also be adjusted in the new data set to improve data quality and data management flexibility.

[0073] For example, assuming that after the semantic similarity calculation of the m items of knowledge data to be processed, m groups of similar semantic data sets are obtained. In the first round of overlap rate calculation, the first similar semantic data set is {q1, q2, q3, ..., q10}, and the second similar semantic data set is {q1, q2, q3, ..., q10}, wherein there are 5 items of overlapping knowledge data in the first similar semantic data set and the second similar semantic data set. Therefore, after executing the first embodiment, the second similar semantic data set is merged into the first similar semantic data set, and the merged first similar semantic data set is {q1, q2, q3, ..., q15}; after executing the second embodiment, a new similar semantic data set is created and the merged result is stored in the third similar semantic data set, namely the third similar semantic data set {q1, q2, q3, ..., q15}. During the second round of overlap rate calculation, the similar semantic data set is updated according to the updated first similar semantic data set or the third similar semantic data set, and the overlap rate between the two data sets is continued to be calculated, and the data set to be merged with the largest overlap rate and greater than the overlap rate threshold is screened out, and the second round of merging operation is continued; finally, the iterative loop is performed until the overlap rate is no greater than the overlap rate threshold to obtain the target knowledge data set.

[0074] In order to improve the accuracy of the target knowledge data set, the present invention provides a method for verifying the target knowledge data to ensure whether there are any overlapping knowledge data items. Figure 1 and Figure 5In the third embodiment of the present invention, the knowledge conflict handling method may further include the following steps:

[0075] Step S41, obtaining answer data corresponding to the target knowledge data set;

[0076] Step S42, performing an overlap check on the target knowledge data set according to the answer data, and determining whether there are any overlapping knowledge data items in the target knowledge data set under the constraint of the answer data;

[0077] If yes, jump to step S43, calculate and merge similar semantic data sets with the largest overlap rate according to the judgment result, and update the target knowledge data set.

[0078] In this way, by verifying the target knowledge data set, the accuracy of knowledge expression can be effectively improved, knowledge redundancy can be reduced, and the quality of the knowledge base can be continuously improved.

[0079] In one embodiment, the target knowledge data set includes some questions in a preset question-and-answer knowledge base. Based on the feedback answer data, it is verified whether the questions themselves are redundant or overlapping, and the quality of the questions themselves is improved by using the feedback of the answer data.

[0080] Specifically, in the first embodiment, if there are identical or similar answer data, it can be further verified whether the corresponding question data is redundant. If there is redundancy, the redundant information can be deleted, and the answer data can also be unified.

[0081] In the second embodiment, it is comprehensively determined whether redundant information exists based on the questions in the target knowledge data set and the corresponding answer data. If redundant information does exist, the redundant information can be deleted and the answer data can be unified.

[0082] In addition, after each round of updating the target knowledge dataset, business experts can evaluate the results to determine whether re-verification is needed, and repeat the iteration multiple times until the evaluation result is that re-verification is not needed to obtain the final target knowledge dataset.

[0083] The above-mentioned various implementation modes, embodiments or specific examples provided by the present invention can be combined with each other to finally form multiple better implementation modes.

[0084] For example, Figure 6 A schematic diagram of the process flow of the knowledge conflict handling method under a preferred implementation mode is shown below. Figure 6 The processing procedure of the preferred embodiment is summarized.

[0085] First, all knowledge data in FAQ are traversed; for each knowledge data, the top 10 knowledge data with a similarity of more than 70% in the remaining knowledge data are retrieved based on the similarity vector retrieval tool Faiss; several groups of semantically similar data sets can be calculated, each of which contains 10 knowledge data.

[0086] Secondly, for each data set x in the several semantically similar data sets, the overlap rate with all other data sets is calculated respectively, and another data set y with the largest overlap rate exceeding 60% is selected, and the two data sets x and y are merged into a new semantically similar data set, and x and y are deleted from the original data set. The operations of calculating the overlap rate and merging data sets are iterated and looped until no data set that can be merged is found, and the target knowledge data set is obtained.

[0087] Finally, the target knowledge data set is verified, and the overlap rate is calculated to determine whether there is still a knowledge conflict problem in the target knowledge data set. If there is still knowledge data that can be merged, the knowledge data with a high overlap rate in the verification result is merged. If there is no knowledge data that can be merged, it is evaluated to determine whether verification is still needed. If verification is not required, the final target knowledge data set is obtained.

[0088] The present invention also provides a refrigeration device, comprising: a memory for storing executable instructions; and a processor for implementing any of the steps of the above-mentioned knowledge conflict processing method when running the executable instructions stored in the memory.

[0089] In summary, the present invention captures the correlation between knowledge data by calculating the semantic similarity between two data, and can group or cluster knowledge data at the semantic level to form multiple groups of similar semantic data sets, which is helpful for the integration of knowledge data, so as to automatically generate multiple groups of knowledge data sets with higher semantic similarity and more aggregation, and provide high-quality data support for subsequent knowledge overlap rate calculation. At the same time, by calculating the overlap rate of two similar semantic data sets and selectively merging knowledge data sets, it is possible to avoid the generation of repeated and redundant knowledge, extract a streamlined and non-repetitive target knowledge data set, and improve the quality of the knowledge data set.

[0090] In addition, the calculation of multiple sets of similar semantic data sets based on semantic similarity and overlap rate is an unsupervised method, which reduces manual intervention to a certain extent and improves processing efficiency and intelligence. At the same time, it can also avoid the errors and costs caused by manual annotation and improve work efficiency.

[0091] It should be understood that although this specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each implementation mode may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0092] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent implementation methods or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

Claims

1. A knowledge conflict handling method, characterized in that: include: Obtain m items of knowledge data to be processed; Calculate the semantic similarity between each knowledge data and the remaining m-1 knowledge data respectively, select the knowledge data with similar semantics to the knowledge data to construct similar semantic data sets, so as to determine m groups of similar semantic data sets; Calculating and selectively merging the knowledge data according to the overlap rate between the m groups of similar semantic data sets to determine a target knowledge data set; wherein the target knowledge data set contains n items of knowledge data without overlap; m≥n.

2. The knowledge conflict handling method according to claim 1, characterized in that: The knowledge data includes all questions in a preset question-and-answer knowledge base; the preset question-and-answer knowledge base includes at least one of a plurality of groups of questions raised by users and a plurality of corresponding groups of answers.

3. The knowledge conflict handling method according to claim 1, characterized in that: The knowledge data is stock problem data.

4. The knowledge conflict handling method according to claim 1, characterized in that: The "respectively calculating the semantic similarity between each piece of knowledge data and the remaining m-1 pieces of knowledge data" specifically includes: Vectorize each piece of knowledge data to obtain the corresponding knowledge vector; Based on a similarity vector retrieval tool, the semantic similarity between each knowledge vector and the remaining m-1 knowledge vectors is calculated to obtain m-1 semantic similarity values; wherein the similarity vector retrieval tool includes Faiss similarity search.

5. The knowledge conflict handling method according to claim 1, characterized in that: The "selecting knowledge data with semantic similarity to the knowledge data to construct a similar semantic data set" specifically includes: According to the m-1 semantic similarity values ​​corresponding to the first knowledge data, all the remaining knowledge data whose semantic similarity values ​​are greater than a preset similarity threshold are judged and determined to form a first reference data set; K items of question data with the highest ranking are obtained by screening the first reference data set and arranging them in descending order of semantic similarity values ​​to form the first similar semantic data set; m≥K.

6. The knowledge conflict handling method according to claim 1, characterized in that: The “calculating and using the overlap ratio between the m groups of similar semantic data sets” specifically includes: Acquire a first similar semantic data set and a second similar semantic data set; Calculating and counting the number of element intersections between the first similar semantic data set and the second similar semantic data set to obtain an amount of overlapping data; comparing the number of elements in the first similar semantic data set and the second similar semantic data set, and counting the number of data sets with a larger number of elements as the merged data amount; The overlap ratio corresponding to the first similar semantic data set and the second similar semantic data set is calculated based on the overlapping data volume and the merged data volume; wherein the overlapping ratio is numerically equal to the ratio of the overlapping data volume to the merged data volume.

7. The knowledge conflict handling method according to claim 1, characterized in that: The "calculating and selectively merging the knowledge data according to the overlap rate between the m groups of similar semantic data sets to determine the target knowledge data set" specifically includes: Calculate the overlap rate of each group of similar semantic data sets with the remaining m-1 groups of similar semantic data sets respectively, and select the similar semantic data set with the maximum overlap rate to determine the data set to be merged; wherein the data set to be merged includes two groups of similar semantic data sets; Determining whether the maximum overlap rate is greater than an overlap rate threshold; If yes, perform a set union operation on the data sets to be merged, and update the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets; Iteratively calculate and selectively merge the intermediate knowledge data sets according to the overlap rates between each two of the intermediate knowledge data sets, until the re-determined maximum overlap rate is not greater than the overlap rate threshold, and the target knowledge data set is determined.

8. The knowledge conflict handling method according to claim 7, characterized in that: The data sets to be merged include a first similar semantic data set and a second similar semantic data set; the "performing a set union operation on the data sets to be merged, and updating the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets" specifically includes: The data elements of the second similar semantic data set are merged into the first similar semantic data set, the first similar semantic data set is updated, and the second similar semantic data set is deleted to obtain the m-1 groups of intermediate knowledge data sets.

9. The knowledge conflict handling method according to claim 7 is characterized in that: The data sets to be merged include a first similar semantic data set and a second similar semantic data set; the "performing a set union operation on the data sets to be merged, and updating the m groups of similar semantic data sets according to the operation result to obtain m-1 groups of intermediate knowledge data sets" specifically includes: Construct an empty new similar semantic dataset, merge the first similar semantic dataset and the second similar semantic dataset, store the merged result in the new similar semantic dataset, and delete the first similar semantic dataset and the second similar semantic dataset to obtain the m-1 groups of intermediate knowledge datasets.

10. The knowledge conflict handling method according to claim 1, characterized in that: After "calculating and selectively merging the knowledge data according to the overlap rate between the m groups of similar semantic data sets to determine the target knowledge data set", the method further includes: Acquire answer data corresponding to the target knowledge data set; According to the answer data, the target knowledge data set is checked for overlap, and it is determined whether there are any overlapping knowledge data items in the target knowledge data set under the constraint of the answer data; If so, based on the judgment result, similar semantic data sets with the largest overlap rate are calculated and merged to update the target knowledge data set.

11. A refrigeration device, characterized in that: include: A memory for storing executable instructions; A processor is used to implement the steps of the knowledge conflict handling method described in any one of claims 1 to 10 when running the executable instructions stored in the memory.

Images