A cross-domain multi-round semantic recognition framework and semantic recognition method

Abstract

The present invention relates to the technical field of natural language processing, and discloses a cross-domain multi-round semantic recognition framework and a semantic recognition method, which aims to solve the problem of poor accuracy of semantic recognition in the existing semantic recognition framework. The scheme mainly includes: a control system, a domain recognition system and a semantic recognition system, the domain recognition system includes a multi-round domain recognition module and a single-round domain recognition module; the control system is used to obtain the corresponding historical dialogue data after obtaining the text to be recognized, and send the text to be recognized and the historical dialogue data to the domain recognition system; and send the text to be recognized, the rewritten text and the domain recognition result to the semantic recognition system; the domain recognition system is used to perform domain recognition, and send the rewritten text and the domain recognition result to the control system, and the semantic recognition system is used to perform semantic recognition, and send the semantic recognition result to the control system. The present invention improves the accuracy of semantic recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a cross-domain multi-round semantic recognition framework and a semantic recognition method. Background Art

[0002] The semantic recognition framework is a set of methods and tools for understanding and analyzing semantic information in text or speech. The semantic recognition framework plays a vital role in natural language processing (NLP), helping computers understand the meaning of text or speech to perform higher-level language processing tasks.

[0003] Most existing semantic recognition frameworks are designed based on specific application scenarios or fields, such as food ordering services, weather queries, etc. Each field is independently modeled and integrated together to identify the field of the text to be recognized by the current user input, and then perform subsequent semantic recognition processing in each field through natural language processing technology. This approach lacks an effective mechanism to coordinate information transmission between different fields, and lacks an effective mechanism to store and use historical conversation data, resulting in limitations in processing cross-domain requests and multi-round conversations. It cannot effectively handle user requests across multiple fields and context dependency issues in multi-round interactions, which in turn makes semantic recognition less accurate. Summary of the invention

[0004] The present invention aims to solve the problem of poor accuracy of semantic recognition in existing semantic recognition frameworks, and proposes a cross-domain multi-round semantic recognition framework and a semantic recognition method.

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

[0006] In a first aspect, the present invention provides a cross-domain multi-round semantic recognition framework, including a control system, a domain recognition system and a semantic recognition system, wherein the domain recognition system includes a multi-round domain recognition module and a single-round domain recognition module;

[0007] The control system is used to obtain the historical conversation data of the corresponding user after obtaining the text to be recognized by the user, and send the text to be recognized and the historical conversation data to the multi-round domain recognition module; and send the text to be recognized and the domain recognition result to the semantic recognition system, and send the rewritten text to the semantic recognition system after receiving it;

[0008] The multi-round domain recognition module is used to determine whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module;

[0009] The single-round domain recognition module is used to perform domain recognition based on the rewritten text or the text to be recognized, and send the corresponding domain recognition result to the control system, and send the rewritten text to the control system when receiving the rewritten text;

[0010] The semantic recognition system is used to perform multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether a rewritten text is received. If so, the text to be recognized is replaced with the rewritten text and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, the text to be recognized is subjected to a single round of caching, rule and algorithm recognition in sequence to obtain the semantic recognition result.

[0011] Furthermore, the multi-round domain identification module includes a multi-round judgment module, a multi-round rewriting module and a rationality judgment module;

[0012] The multi-round judgment module is used to judge whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is sent to the multi-round rewriting module.

[0013] The multi-round rewriting module is used to rewrite the text to be recognized according to the historical conversation data;

[0014] The rationality judgment module is used to judge the rationality of the rewritten text. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module.

[0015] Furthermore, the historical conversation data includes historical text to be recognized, historical rewritten text and historical semantic recognition results.

[0016] Furthermore, the control system is also used to store and update the text to be recognized, the rewritten text and the semantic recognition results by field according to the field recognition results.

[0017] In a second aspect, the present invention provides a semantic recognition method, which is applied to the cross-domain multi-round semantic recognition framework as described in the first aspect, and the method comprises:

[0018] S1. After obtaining the user's text to be recognized, the control system obtains the historical conversation data of the corresponding user, and sends the text to be recognized and the historical conversation data to the multi-round domain recognition module;

[0019] S2. The multi-round domain recognition module determines whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module.

[0020] S3, the single-round domain recognition module performs domain recognition according to the rewritten text or the text to be recognized, and sends the corresponding domain recognition result to the control system, and sends the rewritten text to the control system when receiving the rewritten text;

[0021] S4, the control system sends the text to be recognized and the domain recognition result to the semantic recognition system, and sends the rewritten text to the semantic recognition system after receiving it;

[0022] S5. The semantic recognition system performs multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether the rewritten text is received. If so, the text to be recognized is replaced with the rewritten text, and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, a single round of caching, rule and algorithm recognition is performed on the text to be recognized to obtain the semantic recognition result.

[0023] Furthermore, the historical conversation data includes historical text to be recognized, historical rewritten text and historical semantic recognition results.

[0024] Furthermore, the method also includes: the control system stores and updates the text to be recognized, the rewritten text and the semantic recognition result by field according to the field recognition result.

[0025] The beneficial effects of the present invention are as follows: the cross-domain multi-round semantic recognition framework and semantic recognition method provided by the present invention, in view of the limitations of traditional single-domain dialogue systems in processing cross-domain or multi-round dialogue tasks, add a multi-round domain recognition module to the domain recognition system, use the multi-round domain recognition module to judge the integrity of user intentions, rewrite the text to be recognized according to historical dialogue data, and use the rewritten text for domain recognition and semantic recognition, thereby effectively avoiding the problem of semantic deviation and improving the accuracy of semantic recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the structure of a cross-domain multi-round semantic recognition framework provided by an embodiment of the present invention;

[0027] Figure 2A schematic diagram of the structure of another cross-domain multi-round semantic recognition framework provided by an embodiment of the present invention;

[0028] Figure 3 A flowchart of a semantic recognition method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0030] In some of the processes described in the specification of the present invention and the above-mentioned drawings, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel, and the sequence numbers of the operations are only used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel.

[0031] Since current semantic recognition solutions lack an effective mechanism to store and utilize historical conversation data, they are unable to effectively handle user requests across multiple fields and context dependency issues in multiple rounds of interactions, which results in poor accuracy of semantic recognition.

[0032] Based on this, the technical solution of the present invention is proposed. In an embodiment of the present invention, a cross-domain multi-round semantic recognition framework includes a control system, a domain recognition system and a semantic recognition system. The domain recognition system includes a multi-round domain recognition module and a single-round domain recognition module; the control system is used to obtain the historical conversation data of the corresponding user after obtaining the user's text to be recognized, and send the text to be recognized and the historical conversation data to the multi-round domain recognition module; and send the text to be recognized and the domain recognition result to the semantic recognition system, and send the rewritten text to the semantic recognition system after receiving it; the multi-round domain recognition module is used to determine whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is determined If it is reasonable, the rewritten text will be sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized will be sent to the single-round domain recognition module; the single-round domain recognition module is used to perform domain recognition according to the rewritten text or the text to be recognized, and send the corresponding domain recognition result to the control system, and send the rewritten text to the control system when the rewritten text is received; the semantic recognition system is used to perform multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the domain corresponding to the domain recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether the rewritten text is received. If so, the text to be recognized is replaced with the rewritten text, and single-round caching, rule and algorithm recognition are performed in sequence to obtain the semantic recognition result. If not, the text to be recognized is subjected to single-round caching, rule and algorithm recognition in sequence to obtain the semantic recognition result.

[0033] Specifically, the present invention adds a multi-round domain recognition module to the domain recognition system of the semantic recognition framework. The multi-round domain recognition module first determines whether the intention of the user's text to be recognized is complete. If not, the user's historical conversation data is used to rewrite the text to be recognized, and the rationality of the rewritten text is judged. If it is reasonable, the domain recognition is performed based on the rewritten text and using a single-round domain recognition module. In the process of semantic recognition using the semantic recognition model of the corresponding field, if the multi-round cache and rule recognition of the text to be recognized fail, it is determined whether the rewritten text is received. If so, the rewritten text is used to perform cache, rule and algorithm recognition in sequence to obtain the semantic recognition result. Otherwise, the text to be recognized is sequentially cached, rule and algorithm recognized in sequence to obtain the semantic recognition result. The present invention uses historical conversation data to perform intention judgment, text rewriting and rationality judgment on the text to be recognized, effectively utilizes historical conversation data, can effectively handle user requests across multiple fields and context dependency problems in multi-round interactions, and improves the accuracy of semantic recognition.

[0034] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0035] Figure 1 For a schematic diagram of a cross-domain multi-round semantic recognition framework provided by an embodiment of the present invention, please refer to Figure 1 ,The cross-domain multi-round semantic recognition framework includes a control system, a domain recognition system and a semantic recognition system, wherein the domain recognition system includes a multi-round domain recognition module and a single-round domain recognition module;

[0036] The control system is used to obtain the historical conversation data of the corresponding user after obtaining the text to be recognized by the user, and send the text to be recognized and the historical conversation data to the multi-round domain recognition module; and send the text to be recognized and the domain recognition result to the semantic recognition system, and send the rewritten text to the semantic recognition system after receiving it;

[0037] The multi-round domain recognition module is used to determine whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module;

[0038] The single-round domain recognition module is used to perform domain recognition based on the rewritten text or the text to be recognized, and send the corresponding domain recognition result to the control system, and send the rewritten text to the control system when receiving the rewritten text;

[0039] The semantic recognition system is used to perform multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether a rewritten text is received. If so, the text to be recognized is replaced with the rewritten text and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, the text to be recognized is subjected to a single round of caching, rule and algorithm recognition in sequence to obtain the semantic recognition result.

[0040] See also Figure 2 In an embodiment of the present invention, the multi-round domain identification module and the single-round domain identification module respectively include a cache unit, a rule unit and an algorithm unit, wherein the algorithm unit of the multi-round domain identification module includes a multi-round judgment module, a multi-round rewriting module and a rationality judgment module.

[0041] The multi-round judgment module is used to judge whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is sent to the multi-round rewriting module.

[0042] The multi-round rewriting module is used to rewrite the text to be recognized according to the historical conversation data;

[0043] The rationality judgment module is used to judge the rationality of the rewritten text. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module.

[0044] In the embodiment of the present invention, semantic recognition also includes multi-round recognition and single-round recognition. Similarly, multi-round recognition and single-round recognition also include a cache unit, a rule unit and an algorithm unit, respectively.

[0045] Figure 3 A flowchart of a semantic recognition method is shown, which is applied to the cross-domain multi-round semantic recognition framework described in an embodiment of the present invention. Figure 3 , including the following steps:

[0046] S1. After obtaining the text to be recognized by the user, the control system obtains the historical conversation data of the corresponding user, and sends the text to be recognized and the historical conversation data to the multi-round domain recognition module.

[0047] In actual applications, after the control system obtains the user's text to be recognized, it can obtain the corresponding user's historical conversation data from the database. The historical conversation data may include historical text to be recognized, historical rewritten text, and historical semantic recognition results. Then the text to be recognized and the historical conversation data are transmitted to the domain recognition system.

[0048] S2. The multi-round domain recognition module determines whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical dialogue data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module and the control system. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module.

[0049] After receiving the text to be identified and the historical conversation data, the domain identification system first uses the multi-round domain identification module to perform multiple rounds of domain judgments, and then performs domain identification.

[0050] Specifically, the multi-round domain recognition module uses the cache unit, the rule unit, and the algorithm unit in turn for processing and recognition.

[0051] The cache unit is used for semantic caching, that is, the semantic cache interprets and stores the semantics of the user's text to be recognized, allowing the system to retrieve information based on intent rather than just literal matching. The semantic cache can dynamically retrieve and provide the most relevant data for the current text to be recognized, ensuring that each response meets the user's needs.

[0052] The rule unit is used to apply a series of predefined rules to further process the text to be recognized. These rules can help the system understand the context and intent of the query, so as to provide relevant information more accurately. The rules can include grammatical rules, semantic rules, etc., to ensure that the processing of the text to be recognized conforms to specific logic and standards.

[0053] The algorithm unit includes a multi-round judgment module, a multi-round rewriting module and a rationality judgment module. Each unit or module can be implemented in different ways and is scalable. The multi-round judgment module is responsible for judging whether the intention of the user's text to be recognized in this round is complete, that is, whether historical dialogue data is needed to understand. If the intention is complete, historical dialogue data is not required, and the text to be recognized is directly sent to the single-round domain recognition module for domain recognition; if the intention is incomplete, historical dialogue data is required, and the text to be recognized is sent to the multi-round rewriting module. The multi-round rewriting module rewrites the text to be recognized in this round according to the historical dialogue data transmitted by the control system, outputs the rewritten text with the user's complete intention, and enters the rationality judgment module. The rationality judgment module is an optional execution module used to judge whether the rewritten text is reasonable. It belongs to the fault-tolerant module of the multi-round rewriting module. If it is unreasonable, the rewriting fails. At this time, the text to be recognized is sent to the single-round domain recognition module for domain recognition; if it is reasonable, the rewriting is successful. At this time, the rewritten text is sent to the single-round domain recognition module for domain recognition.

[0054] For example, in the user's historical conversation data, the text to be recognized in the previous round is "Is Chengdu fun?", and the text to be recognized in this round is "Will the weather be good there tomorrow?". The multi-round judgment module determines whether the intention of the user's text to be recognized in this round is complete. The intention of "Will the weather be good there tomorrow?" is incomplete, and the text to be recognized in this round is passed to the multi-round rewriting module. The multi-round rewriting module is based on the historical conversation data. If the multi-round rewriting module rewrites the text of "Will the weather be good there tomorrow?" to "Will the weather in Chengdu be good tomorrow?", the rationality judgment module determines that "Will the weather be good in Chengdu tomorrow?" is reasonable, and at this time "Will the weather be good in Chengdu tomorrow?" is sent to the single-round domain recognition module for domain recognition. If the multi-round rewriting module rewrites the text of "Will the weather be good there tomorrow?" to "Will the weather be good in Chengdu?", the rationality judgment module determines that "Will the weather be good in Chengdu tomorrow?" is unreasonable, and at this time "Will the weather be good there tomorrow?" is sent to the single-round domain recognition module for domain recognition, and the user may need to complete the text to be recognized later.

[0055] Among them, the multi-round judgment module can determine whether multiple rounds of rewriting are needed by analyzing whether the text to be recognized is complete, and the specific analysis methods include but are not limited to text classification and other methods. The multi-round rewriting module rewrites the text to be recognized in this round according to the user's historical dialogue data transmitted by the control system, and the specific rewriting methods include but are not limited to text generation and other methods. The rationality judgment module judges whether the rewritten text is reasonable, and the specific judgment methods include but are not limited to text classification and other methods.

[0056] S3, the single-round domain recognition module performs domain recognition according to the rewritten text or the text to be recognized, and sends the corresponding domain recognition result to the control system, and sends the rewritten text to the control system when receiving the rewritten text;

[0057] In actual applications, after receiving the rewritten text, the single-round domain recognition module performs domain recognition based on the received rewritten text. After receiving the text to be recognized but not the rewritten text, the single-round domain recognition module performs domain recognition based on the text to be recognized. Specifically, the cache unit, the rule unit and the algorithm unit are used in sequence to process and recognize the rewritten text or the text to be recognized to obtain the domain recognition result.

[0058] Specifically, the cache unit and rule unit of the single-round domain recognition module are processed and recognized in the same manner as the cache unit and rule unit of the multi-round domain recognition module, which will not be repeated here. The algorithm unit of the single-round domain recognition module uses various algorithms to process and recognize the text to be recognized or rewritten. These algorithms may include natural language processing (NLP) algorithms, machine learning algorithms, etc., which are used to analyze the language features of the query, sentiment analysis, entity recognition and other tasks.

[0059] The single-round domain recognition module finally sends the domain recognition results obtained by recognition to the control system. If there is a rewritten text, the rewritten text will also be sent to the control system.

[0060] S4. The control system sends the text to be recognized and the domain recognition result to the semantic recognition system, and sends the rewritten text to the semantic recognition system after receiving it.

[0061] In actual applications, after receiving the domain recognition result, the control system transmits the text to be recognized and its corresponding domain recognition result to the semantic recognition system for semantic recognition. If the control system also receives the rewritten text, the rewritten text is also transmitted to the semantic recognition system.

[0062] S5. The semantic recognition system performs multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether the rewritten text is received. If so, the text to be recognized is replaced with the rewritten text, and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, a single round of caching, rule and algorithm recognition is performed on the text to be recognized to obtain the semantic recognition result.

[0063] See also Figure 2 The semantic recognition system also includes multiple rounds of recognition and multiple rounds of recognition processes. The semantic recognition system first determines the corresponding semantic recognition model based on the domain recognition, and then uses the semantic recognition model for semantic recognition. For example, the text to be recognized in this round is "Will the weather be good there tomorrow?", and its corresponding domain recognition result is the weather domain, then the semantic recognition model corresponding to the weather domain is used for semantic recognition.

[0064] When performing semantic recognition, the semantic recognition model first performs multiple rounds of recognition and then performs single round recognition. Specifically, during the multiple rounds of recognition, the semantic recognition model first uses the cache unit and the rule unit to perform multiple rounds of caching and rule recognition on the text to be recognized. If the recognition is successful, the semantic recognition is completed, and the semantic recognition result corresponding to the text to be recognized is obtained, and then the semantic recognition result is returned to the control system. If the recognition fails, it is determined whether the rewritten text is received. If so, the text replacement unit is used to replace the text to be recognized with the rewritten text, and then a single round of recognition is entered. The cache unit, the rule unit, and the algorithm unit are used to cache, recognize rules and algorithms on the rewritten text in turn, and obtain the semantic recognition result corresponding to the rewritten text. If not, the cache unit, the rule unit, and the algorithm unit are used to perform single rounds of caching, rules and algorithm recognition on the text to be recognized in turn to obtain the semantic recognition result. Finally, the semantic recognition result is returned to the control system.

[0065] In the embodiment of the present invention, the control system stores and updates the text to be recognized, the rewritten text and the semantic recognition results by field according to the field recognition results, and generates historical conversation data corresponding to different fields of the user for the next call.

[0066] In summary, the cross-domain multi-round semantic recognition framework and semantic recognition method provided by the embodiment of the present invention add a multi-round domain recognition module to the domain recognition system of the semantic recognition framework. The multi-round domain recognition module first determines whether the intention of the user's text to be recognized is complete. If not, the user's historical conversation data is used to rewrite the text to be recognized, and the rationality of the rewritten text is judged. If it is reasonable, the domain recognition is performed based on the rewritten text and using a single-round domain recognition module. In the process of semantic recognition using the semantic recognition model of the corresponding field, if the multi-round cache and rule recognition of the text to be recognized fail, the rewritten text is used to perform single-round cache, rule and algorithm recognition in sequence to obtain the semantic recognition result. The embodiment of the present invention uses historical conversation data to judge the intention, rewrite the text and judge the rationality of the text to be recognized, effectively utilizes historical conversation data, can effectively handle user requests across multiple fields and context dependency problems in multi-round interactions, and improves the accuracy of semantic recognition.

Claims

1. A cross-domain multi-round semantic recognition framework, characterized in that: It includes a control system, a domain recognition system and a semantic recognition system, wherein the domain recognition system includes a multi-round domain recognition module and a single-round domain recognition module; The control system is used to obtain the historical conversation data of the corresponding user after obtaining the text to be recognized by the user, and send the text to be recognized and the historical conversation data to the multi-round domain recognition module; and sending the text to be recognized and the domain recognition result to the semantic recognition system, and sending the rewritten text to the semantic recognition system after receiving it; The multi-round domain recognition module is used to determine whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module; The single-round domain recognition module is used to perform domain recognition based on the rewritten text or the text to be recognized, and send the corresponding domain recognition result to the control system, and send the rewritten text to the control system when receiving the rewritten text; The semantic recognition system is used to perform multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether a rewritten text is received. If so, the text to be recognized is replaced with the rewritten text and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, the text to be recognized is subjected to a single round of caching, rule and algorithm recognition in sequence to obtain the semantic recognition result.

2. The cross-domain multi-round semantic recognition framework according to claim 1 is characterized in that: The multi-round domain identification module includes a multi-round judgment module, a multi-round rewriting module and a rationality judgment module; The multi-round judgment module is used to judge whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is sent to the multi-round rewriting module. The multi-round rewriting module is used to rewrite the text to be recognized according to the historical conversation data; The rationality judgment module is used to judge the rationality of the rewritten text. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module.

3. The cross-domain multi-round semantic recognition framework according to claim 1 is characterized in that: The historical conversation data includes historical text to be recognized, historical rewritten text and historical semantic recognition results.

4. The cross-domain multi-round semantic recognition framework according to claim 1 is characterized in that: The control system is also used to store and update the text to be recognized, the rewritten text and the semantic recognition results by field according to the field recognition results.

5. A semantic recognition method, characterized in that: Applied to the cross-domain multi-round semantic recognition framework as claimed in any one of claims 1 to 4, the method comprises: S1. After obtaining the user's text to be recognized, the control system obtains the historical conversation data of the corresponding user, and sends the text to be recognized and the historical conversation data to the multi-round domain recognition module; S2. The multi-round domain recognition module determines whether the intention of the text to be recognized is complete. If so, the text to be recognized is sent to the single-round domain recognition module. If not, the text to be recognized is rewritten according to the historical conversation data, and the rationality of the rewritten text is judged. If it is judged to be reasonable, the rewritten text is sent to the single-round domain recognition module. If it is judged to be unreasonable, the text to be recognized is sent to the single-round domain recognition module. S3, the single-round domain recognition module performs domain recognition according to the rewritten text or the text to be recognized, and sends the corresponding domain recognition result to the control system, and sends the rewritten text to the control system when receiving the rewritten text; S4, the control system sends the text to be recognized and the domain recognition result to the semantic recognition system, and sends the rewritten text to the semantic recognition system after receiving it; S5. The semantic recognition system performs multiple rounds of caching and rule recognition on the text to be recognized according to the semantic recognition model of the field corresponding to the field recognition result. If the recognition is successful, the semantic recognition result is obtained. If the recognition fails, it is determined whether the rewritten text is received. If so, the text to be recognized is replaced with the rewritten text, and a single round of caching, rule and algorithm recognition is performed in sequence to obtain the semantic recognition result. If not, a single round of caching, rule and algorithm recognition is performed on the text to be recognized to obtain the semantic recognition result.

6. The semantic recognition method according to claim 5, characterized in that: The historical conversation data includes historical text to be recognized, historical rewritten text and historical semantic recognition results.

7. The semantic recognition method according to claim 5, characterized in that: The method further comprises: the control system stores and updates the text to be recognized, the rewritten text and the semantic recognition result by field according to the field recognition result.

Images