Dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion

Abstract

This invention discloses a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion. This invention relates to the field of data interaction technology and solves the problem that existing technologies lack a complete semantic quantification and comprehensive comparison mechanism. Most solutions only perform text clustering through single-dimensional features and fail to construct a quantifiable semantic representation system. This invention uses the main sequence as a benchmark to perform mean proportion calculations on all vectors within the linked sequence set, while also considering the comprehensive similarity of multiple vector groups within the same sequence set, avoiding classification bias caused by single comparisons. Through a dual labeling method of calibrating classification features and extracting individual features, it achieves accurate clustering of semantically similar texts while preserving the unique attributes of individual texts. This ensures that data classification satisfies both group semantic consistency and individual differences, significantly reducing the probability of cross-category mixing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data interaction, in particular to a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion. BACKGROUND

[0002] In the context of explosive growth of heterogeneous data, the data classification and management capability of the knowledge base, as the core carrier of information storage and intelligent interaction, directly determines the interaction efficiency and accuracy.

[0003] However, in the existing background technology, the knowledge base data processing scheme generally has many bottlenecks, which is difficult to adapt to the efficient management and control needs of complex heterogeneous texts.

[0004] Traditional knowledge bases mostly use keyword matching or manual classification methods. The former can only compare the surface text features and cannot deeply mine the semantic association behind the text, which is prone to classification bias due to keyword ambiguity or absence, and has a high probability of cross-classification mixing. The latter relies on manual classification and labeling of texts one by one, which is time-consuming, labor-intensive, and costly, and is also affected by subjective judgment, making it difficult to ensure the consistency of classification standards, and the efficiency is extremely low in large-scale text processing scenarios, which cannot meet the real-time needs.

[0005] At the same time, the existing technology lacks a perfect semantic quantification and comprehensive comparison mechanism. Most schemes only use single-dimensional features for text clustering, without building a quantifiable semantic representation system, making it difficult to accurately measure the semantic similarity between different texts, and the comparison process mostly uses single comparison logic, ignoring the collaborative association between multiple texts, resulting in insufficient logic and accuracy of the classification results.

[0006] In addition, the updating mechanism of traditional knowledge bases is relatively rigid. When new texts are accessed, it is difficult to quickly match the existing classification system, resulting in semantic distortion or disordered storage, making it difficult to achieve dynamic adaptation and sustainable updating. Moreover, the core algorithm and parameters of most systems are fixed, making it difficult to adapt to the needs of heterogeneous text processing in different fields such as finance, education, and medicine, limiting the scalability and practicality.

[0007] Under this background, there is an urgent need for an intelligent interaction system that can accurately represent semantics, automatically and efficiently classify, and dynamically adapt and update to solve the pain points of existing technology. SUMMARY

[0008] To overcome the shortcomings of the prior art, the present application provides a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion, which solves the problem of lack of perfect semantic quantification and comprehensive comparison mechanism in the prior art, and most schemes only use single-dimensional features for text clustering, without building a quantifiable semantic representation system.

[0009] To achieve the above object, the application is implemented by the following technical solutions: a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion, comprising:

[0010] A semantic vector confirmation end, which confirms vectors for different text contents in the semantic partition database, identifies semantic vectors associated with different text contents, and transmits the identified semantic vectors to the partition adjustment processing end, in a specific manner as follows:

[0011] The semantic vector confirmation end confirms the text title associated with the corresponding text content, sorts the relevant characters associated in the text title to generate a character sorting column, confirms the semantic vector associated with the corresponding character based on a preset character semantic comparison table, and sorts the generated semantic vector to generate a semantic vector sorting column, wherein the character semantic comparison table is a preset table.

[0012] The semantic vector confirmation end determines different semantic vector sorting columns associated with different text contents and transmits the sequentially determined semantic vector sorting columns to the partition adjustment processing end.

[0013] The partition adjustment processing end randomly combines multiple different text contents based on different semantic vector sorting columns associated with different text contents, divides the content combination that meets the similarity standard into a sequence set during the combination process, divides the text content associated therewith into the same group classification, and synchronously locks the same semantic vector associated with the corresponding classification as the classification feature associated with the corresponding classification, in a specific manner as follows:

[0014] From the semantic vector sorting columns associated with a plurality of groups of text contents, a group of semantic vector sorting columns is randomly selected as a master sequence, and a master-slave sequence with a similarity of 80% or more than the master sequence is searched from the remaining plurality of groups of semantic vector sorting columns, wherein the similarity is confirmed in the following manner: the semantic vectors at the same sorting position of the master sequence and other semantic vector sorting columns are compared to identify whether the semantic vectors at the same sorting position are the same, if they are the same, they are recorded as the same type of vector, otherwise, no marking is performed, the proportion of vectors of the same type of vector located in the master sequence and the proportion of vectors located in other semantic vector sorting columns are confirmed, the two groups of vector proportions are processed by averaging, the proportion average is confirmed as the confirmed similarity, the master sequence and the master-slave sequence are recorded as a sequence set, and a master-slave sequence with a similarity of 80% or more than the sequence set is searched from the plurality of groups of semantic vector sorting columns, if it exists, the searched master-slave sequence is synchronously placed in the sequence set, if it does not exist, the master sequence is randomly selected from the remaining semantic vector sorting columns, and the confirmation process of other sequence sets is performed.

[0015] The determination of the master-slave sequence in the sequence set is: a group of other semantic vector order columns is selected from the remaining semantic vector order columns, and the similarity is confirmed with the semantic vectors existing in the sequence set in turn, and the confirmed multiple sets of similarity are processed by averaging to confirm the comprehensive similarity. If the comprehensive similarity exceeds 80%, it means that the other semantic vector order column is located in this sequence set, and the semantic vector is the determined master-slave sequence. Otherwise, it means that it is not located in this sequence set;

[0016] The overall confirmation process of the sequence set is performed on a plurality of groups of semantic vector order columns, and the semantic vector columns belonging to the same sequence set are divided into the same semantic vector column, and the text content associated in the semantic partition database is classified;

[0017] The same semantic vectors associated at the same sorting position of the same sequence set are marked, and sorted in the same order before and after, as the classification features associated with the corresponding classification partition;

[0018] The single feature associated with each text content is marked again, the semantic vector column associated with the text content is confirmed, the semantic vector not belonging to the classification feature is extracted according to the unchanged order before and after, a group of vector columns are generated, and the sorting position of different semantic vectors in the original semantic vector order column is confirmed. The position mark of the corresponding semantic vector is generated to generate the single feature associated with the corresponding text content;

[0019] The dynamic update processing end, in the text content input process, confirms the semantic vector order column associated with the input text content through the semantic vector confirmation end, and identifies the associated partition of the text content according to the classification features of different classification partitions in the semantic partition database. If there is only one group in the associated partition, it is directly divided into the classification partition. If there are multiple groups in the associated partition, the single features of different text contents in different classification partitions are identified, the high similarity objects are confirmed, and they are divided into the corresponding classification partition. The specific way is:

[0020] Confirm the semantic vector order column associated with the input text content, and confirm the classification features of different classification partitions simultaneously. Identify the similar features of the classification features and the semantic vector order column: identify whether the semantic vectors associated with the classification features exist in the semantic vector order column:

[0021] If it exists, the same semantic vector associated with the semantic vector order column is extracted, and the extracted same semantic vector is sorted according to the relationship before and after to generate a group of to-be-verified sequences. Then, it is identified whether the to-be-verified sequence is completely consistent with the classification feature. If it is completely consistent, the classification partition is marked as the associated partition. If it is not completely consistent, no marking is performed;

[0022] If not, no calibration is performed;

[0023] If the associated partition only has one group, the text content is directly divided into the associated partition for storage;

[0024] If the associated partition has multiple groups, the text content is not divided into other semantic vectors of the to-be-verified sequence, the sorting position is recorded, the position vector sequence is generated, and it is confirmed whether there is a monomer feature with the same position relationship as the position vector sequence from the monomer features of different text contents associated with different associated partitions. If there is, the monomer feature is recorded as a pending feature, otherwise, the text content is randomly divided into any one of the multiple associated partitions for storage;

[0025] If the pending feature only has one group, the text content is directly divided into the associated partition for storage;

[0026] If the pending feature has multiple groups, the similarity of the position vector sequence and the pending feature is identified, the monomer feature with the highest similarity is selected as the determined feature, and the text content is directly divided into the associated partition corresponding to the determined feature for storage, and the dynamic updating process of the corresponding partition database is completed.

[0027] The application provides a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion. Compared with the prior art, the following beneficial effects are achieved:

[0028] The partition adjustment processing end adopts a multi-round comparison mechanism with an 80% similarity threshold, calculates the average proportion of all vector sorting columns in the main sequence as the benchmark linkage sequence set, and simultaneously considers the comprehensive similarity of multiple vectors in the same sequence set, thereby avoiding classification deviation caused by single comparison; through the double marking mode of calibration classification features and extracted monomer features, accurate clustering of texts with the same semantics is realized, and unique properties of single texts are retained, so that data classification not only meets group semantic consistency, but also considers individual differences, thereby greatly reducing the cross-classification mixing probability;

[0029] The random master selection and multi-round expansion sequence set construction mode of the partition adjustment processing end does not need to preset a classification framework, can adapt to the semantic distribution characteristics of text data, and quickly completes the grouping and clustering of a large amount of text; the entire classification process is promoted through an automatic algorithm of vector comparison, without manually judging and classifying the semantics of text content one by one, thereby significantly improving the efficiency of large-scale data processing, and being especially suitable for quick sorting and optimization of massive texts in a knowledge base;

[0030] By employing a two-layer verification logic that first matches classification features and then compares individual features, accurate partition classification of newly added text is achieved: text with a single associated partition is directly classified, text with multiple associated partitions is classified by comparing the similarity between the position vector sequence and the individual features, and text without a matching partition is temporarily stored for processing. This ensures the compatibility of the new data with the existing knowledge base classification system and avoids semantic distortion caused by forced classification. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the principle framework of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] First Embodiment

[0034] Please see Figure 1 This application provides a dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion, including a semantic vector confirmation end, a semantic partition database, a partition adjustment processing end, and a dynamic update processing end. The semantic vector confirmation end, the partition adjustment processing end, and the semantic partition database are electrically connected from the output node to the input node in sequence, and the semantic partition database is electrically connected to the input node of the semantic vector confirmation end. The semantic vector confirmation end, the dynamic update processing end, and the semantic partition database are electrically connected from the output node to the input node in sequence.

[0035] Among them, the semantic partitioning database stores different text content, and its text data is provided in advance by relevant external personnel;

[0036] The semantic vector confirmation end performs vector confirmation on different text contents within the semantic partitioning database, identifies the semantic vectors associated with different text contents, and transmits the identified semantic vectors to the partitioning adjustment processing end. The specific method for identifying semantic vectors is as follows:

[0037] Identify the text title associated with the corresponding text content, sort the related characters in the text title to generate a character sorting sequence, and then, based on a preset character semantic comparison table, identify the semantic vector associated with the corresponding character, sort the generated semantic vectors to generate a semantic vector sorting sequence. The character semantic comparison table is a preset table, which is prepared in advance by relevant personnel based on experience.

[0038] The different semantic vector sequences associated with different text contents are determined, and the determined semantic vector sequences are transmitted to the partition adjustment processing terminal.

[0039] Specifically, each different text content is associated with a different text title. Based on the related characters associated with the text title, the semantic vector sequence associated with the sorted characters can be effectively determined. This allows each different text content to be associated with a set of semantic vector sequences, facilitating subsequent classification of different text content and dividing several different text contents into multiple groups of different text contents of the same category.

[0040] Among them, the partition adjustment processing end sorts the different semantic vectors associated with different text content, randomly combines multiple different text content, and divides the content that meets the similarity standard into a sequence set during the combination process, and classifies the associated text content into the same category, and simultaneously locks the same semantic vector associated with the corresponding category as the classification feature associated with the corresponding category.

[0041] The specific method for dividing the sequence set is as follows:

[0042] From a series of semantic vector ranking sequences associated with several sets of text content, a random sequence is selected as the main sequence. From the remaining semantic vector ranking sequences, a master-slave sequence with a similarity of 80% or higher to the main sequence is searched. The similarity is confirmed by comparing semantic vectors in the same ranking position in the main sequence with those in other semantic vector ranking sequences. If they are identical, they are marked as vectors of the same type; otherwise, no marking is made. The proportion of vectors of the same type located in the main sequence and in other semantic vector ranking sequences is then confirmed. The average proportions of the two sets of proportions are calculated, and this average proportion is the confirmed similarity. The main sequence and the master-slave sequence are then compared. The sequence is denoted as a sequence set. Then, from several sets of semantic vector sorted sequences, a master-slave sequence with a similarity of 80% or more is searched for. If it exists, the master-slave sequence is placed in this sequence set. If it does not exist, a master sequence is randomly selected from the remaining semantic vector sorted sequences, and the process of confirming other sequence sets is carried out. For example, if the text content included in the partitioned database is planned to be 10 sets, each corresponding to a different semantic vector sorted sequence, a sorted sequence is randomly selected as the master sequence, and then a master-slave sequence is searched from the remaining 9 sets. The master-slave sequence must meet the requirement that the similarity reaches 80% or more, that is, the average proportion of semantic vectors associated with the same sorted position must meet the standard. The specific sorted sequence that does not meet the standard is not a master-slave sequence of the master sequence.

[0043] The method for determining whether a master-slave sequence is located in a sequence set is as follows: From the remaining semantic vector sequences, select another set of semantic vector sequences and sequentially check their similarity with the semantic vectors already present in the sequence set. Then, average the confirmed similarities to determine the overall similarity. If the overall similarity exceeds 80%, it means the other semantic vector sequence is located within this sequence set, and this semantic vector is the determined master-slave sequence. Conversely, if the similarity is below 80%, it means it is not located within this sequence set. Specifically, the similarity confirmation method is as described above. The result of vector proportion mean processing is as follows: For example, there are two sets of semantic vector sorted sequences L1 and L2 in a sequence set. Then, a set L3 is randomly selected from the remaining sorted sequences. If the similarity between L1 and L3 meets the standard, and the similarity between L2 and L3 does not meet the standard, but the mean of the two sets of similarity meets the standard, then L3 can also be classified into this sequence set. If the similarity of both sets meets the standard, then it can also be classified into this sequence set. If the similarity of both sets does not meet the standard, then it means that L3 cannot be classified into the same category.

[0044] The process involves sequentially confirming the overall sequence set of several groups of semantic vector sorted sequences, dividing semantic vector columns belonging to the same sequence set into semantic vector columns, classifying the associated text content within the semantic partition database, and confirming the classification features of the corresponding classification partitions after completing the classification process.

[0045] The same semantic vectors associated with several groups of semantic vector columns in the same sequence set at the same sorting position are labeled and sorted according to the same sorting method before and after, and used as the classification features associated with the corresponding classification partition.

[0046] Then, the individual features associated with each text content are labeled simultaneously to confirm the semantic vector column associated with the text content. Semantic vectors that do not belong to the classification features are extracted in the same order as before, generating a set of vector columns. The order position of different semantic vectors in the original semantic vector column is confirmed, and position marks are assigned to the corresponding semantic vectors to generate the individual features associated with the corresponding text content. For example, if a set of semantic vector columns associated with a set of text content is ABCDE, where BDE belongs to the classification features of the corresponding category, then the semantic vectors that do not belong to the classification features include A and C, which are associated with order positions 1 and 3 respectively. Then the individual features associated with the corresponding text content can be represented as: A1C3. This is just one representation method, and other methods can be used instead, such as the form without subscripts (A1C3), etc.

[0047] Second Embodiment

[0048] In the specific implementation process of this embodiment, compared with the above embodiment, embodiment one mainly focuses on the specific division process of similar content in the partition database, while this embodiment mainly focuses on the dynamic update process of external text content belonging to the semantic partition database, and its specific execution end is the dynamic update processing end.

[0049] Among them, the dynamic update processing end confirms the semantic vector sequence associated with the input text content through the semantic vector confirmation end during the text content input process. Then, based on the classification features of different classification partitions in the semantic partition database, it identifies the associated partitions of the text content. If there is only one set of associated partitions, it is directly assigned to this classification partition. If there are multiple sets of associated partitions, it identifies the individual features of different text content in different classification partitions, confirms highly similar objects, and assigns them to the corresponding classification partitions.

[0050] The specific method for identifying related sections of text content is as follows:

[0051] Confirm the semantic vector ranking sequence associated with the input text content, and simultaneously confirm the classification features of different classification partitions, identifying similarities between the classification features and the semantic vector ranking sequence: Determine whether all semantic vectors associated with the classification features exist within the semantic vector ranking sequence.

[0052] If they exist, the semantic vectors associated with the semantic vector sorting sequence are extracted, and the extracted semantic vectors are sorted according to their order to generate a set of sequences to be verified. Then, it is identified whether the sequences to be verified are completely consistent with the classification features. If they are completely consistent, this classification partition is marked as an associated partition. If they are not completely consistent, no marking is performed, and the sequences are directly input into the semantic partition database for storage without performing specific classification (that is, they cannot be classified and the similarity does not meet the standard).

[0053] If it does not exist, no labeling is performed, and it is directly input into the semantic partitioning database for storage without specific classification (that is, it cannot be classified and the similarity does not meet the standard).

[0054] Identify the total number of associated partitions identified:

[0055] If there is only one associated partition, the text content will be directly allocated to this associated partition for storage.

[0056] If there are multiple sets of associated partitions, the semantic vectors of the text content that have not been divided into the sequence to be verified are confirmed and the sorting position is recorded. A position vector sequence is generated. From the individual features of different text content associated with different associated partitions, it is confirmed whether there are individual features with the same positional relationship as this position vector sequence. If there are, such individual features are recorded as undetermined features. Otherwise, the text content is randomly divided into any one of the multiple sets of associated partitions for storage.

[0057] If there is only one set of features to be determined, then the text content will be directly assigned to this associated partition for storage.

[0058] If there are multiple sets of features to be determined, the similarity between the location vector sequence and the features to be determined is identified (the same as the similarity confirmation method disclosed above). The single feature with the highest similarity is selected as the determined feature, and the text content is directly divided into the associated partition corresponding to the determined feature for storage (if the similarity is the highest and there are the same values, a set of associated partitions can be randomly selected for storage), thus completing the dynamic update process of the corresponding partition database.

[0059] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0060] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A dynamic multi-knowledge base intelligent interactive system based on heterogeneous data fusion, characterized in that, include: The semantic vector confirmation end performs vector confirmation on different text contents in the semantic partitioning database, identifies the semantic vectors associated with different text contents, and transmits the identified semantic vectors to the partitioning adjustment processing end. The partitioning adjustment processing end sorts different semantic vectors associated with different text content, randomly combines multiple different text content, and divides the content that meets the similarity standard into a sequence set during the combination process, and classifies the associated text content into the same category. Simultaneously, it locks the same semantic vector associated with the corresponding category as the classification feature associated with the corresponding category. The dynamic update processing end confirms the semantic vector sequence associated with the input text content through the semantic vector confirmation end during the text content input process. Then, based on the classification features of different classification partitions in the semantic partition database, it identifies the associated partitions of the text content. If there is only one set of associated partitions, it is directly assigned to this classification partition. If there are multiple sets of associated partitions, it identifies the individual features of different text content in different classification partitions, confirms highly similar objects, and assigns them to the corresponding classification partitions.

2. The dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion according to claim 1, characterized in that, The semantic vector confirmation terminal identifies the semantic vectors corresponding to different text contents in the following specific way: Identify the text title associated with the corresponding text content, sort the related characters associated with the text title to generate a character sorting sequence, and then, based on a preset character semantic comparison table, identify the semantic vector associated with the corresponding character, sort the generated semantic vectors to generate a semantic vector sorting sequence, where the character semantic comparison table is a preset table; The different semantic vector sequences associated with different text contents are determined, and the determined semantic vector sequences are transmitted to the partition adjustment processing terminal.

3. The dynamic multi-knowledge base intelligent interactive system based on heterogeneous data fusion according to claim 1, characterized in that, The partitioning adjustment processing terminal divides the sequence set in the following specific way: From a number of semantic vector sequences associated with text content, a set of semantic vector sequences is randomly selected and designated as the main sequence. From the remaining sets of semantic vector sequences, master-slave sequences with a similarity of 80% or more to the main sequence are searched. The similarity is confirmed by comparing the semantic vectors in the same sorting position in the main sequence with those in other semantic vector sequences. If they are the same, they are marked as vectors of the same type; otherwise, no marking is made. The proportion of vectors of the same type in the main sequence and in other semantic vector sequences is then confirmed. The proportions of the two sets of vectors are averaged to confirm the average proportion, which is the confirmed similarity. The main sequence and master-slave sequences are recorded as a sequence set. Then, master-slave sequences that meet the similarity standard of this sequence set are searched from the remaining sets of semantic vector sequences. If they exist, the searched master-slave sequences are placed in this sequence set. If they do not exist, a new main sequence is randomly selected from the remaining sets of semantic vector sequences, and the process of confirming other sequence sets is repeated. The method for identifying the master-slave sequence in the sequence set is as follows: Select another set of semantic vector sequences from the remaining semantic vector sequences, and check their similarity with the semantic vectors existing in the sequence set in turn. Then, average the confirmed similarity of multiple sets to confirm the comprehensive similarity. If the comprehensive similarity exceeds 80%, it means that the other semantic vector sequence is in this sequence set, and this semantic vector is the determined master-slave sequence. Otherwise, it means that it is not in this sequence set. The process involves sequentially confirming the sequence set of several groups of semantic vector sorting sequences, dividing semantic vector columns belonging to the same sequence set into the same semantic vector column, and classifying the associated text content within the semantic partition database.

4. The dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion according to claim 3, characterized in that, The specific method by which the partition adjustment processing terminal determines the classification features of the classification partitions is as follows: The same semantic vectors associated with several groups of semantic vector columns in the same sequence set at the same sorting position are labeled and sorted according to the same sorting method before and after, and used as the classification features associated with the corresponding classification partition. Then, the individual features associated with each text content are labeled simultaneously, the semantic vector columns associated with the text content are confirmed, the semantic vectors that do not belong to the classification features are extracted in the same order, a set of vector columns are generated, and the sorting position of different semantic vectors in the original semantic vector sorting column is confirmed, the position mark of the corresponding semantic vector is assigned, and the individual features associated with the corresponding text content are generated.

5. The dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion according to claim 1, characterized in that, The specific method by which the dynamic update processing terminal identifies the associated partitions of text content is as follows: Confirm the semantic vector ranking sequence associated with the input text content, and simultaneously confirm the classification features of different classification partitions, identifying similarities between the classification features and the semantic vector ranking sequence: Determine whether all semantic vectors associated with the classification features exist within the semantic vector ranking sequence. If they exist, the semantic vectors associated with the semantic vector sorting sequence are extracted, and the extracted semantic vectors are sorted according to their order to generate a set of sequences to be verified. Then, it is identified whether the sequences to be verified are completely consistent with the classification features. If they are completely consistent, this classification partition is marked as an associated partition. If they are not completely consistent, no marking is performed. If it does not exist, no calibration is performed.

6. The dynamic multi-knowledge base intelligent interaction system based on heterogeneous data fusion according to claim 5, characterized in that, The dynamic update processing terminal executes different processing methods based on the different total numbers of associated partitions: If there is only one associated partition, the text content will be directly allocated to this associated partition for storage. If there are multiple sets of associated partitions, the semantic vectors of the text content that have not been divided into the sequence to be verified are confirmed and the sorting position is recorded to generate a position vector sequence. From the individual features of different text content associated with different associated partitions, it is confirmed whether there are individual features with the same positional relationship as this position vector sequence. If there are, such individual features are recorded as undetermined features. Otherwise, the text content is randomly divided into any one of the multiple sets of associated partitions for storage.

7. The dynamic multi-knowledge base intelligent interactive system based on heterogeneous data fusion according to claim 6, characterized in that, The dynamic update processing terminal performs different processing methods based on the different total numbers of the undetermined features: If there is only one set of features to be determined, then the text content will be directly assigned to this associated partition for storage. If there are multiple sets of features to be determined, the similarity between the position vector sequence and the features to be determined is identified, the single feature with the highest similarity is selected as the determined feature, and the text content is directly divided into the associated partition corresponding to the determined feature for storage, thus completing the dynamic update process of the corresponding partition database.

Images