Artificial intelligence entity, voice information recognition method, storage medium and program product

Abstract

The present application relates to an artificial intelligence entity, a method for recognizing speech information, a storage medium and a program product. The artificial intelligence entity includes a memory and a processor, wherein a computer program is stored in the memory, and the processor is used to implement the following steps when calling and executing the computer program: in response to the target speech information, the semantic information of the target speech information is recognized by a preset semantic recognition model; according to the target speech information and the semantic information, the response information is determined, and the response information is output. Among them, the semantic recognition model is obtained by training with large sample annotated data, and the large sample annotated data is obtained by data expansion processing of the small sample annotated data through a pre-constructed semantic knowledge graph, and the semantic knowledge graph includes corpora in a variety of human-computer interaction scenarios. In this way, the semantic information of the target speech information can be accurately identified through the semantic recognition model, and then the response information that is more in line with the user's interaction expectations can be output, and the human-computer interaction effect is better.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to an artificial intelligence entity, a voice information recognition method, a storage medium and a program product. Background Art

[0002] With the rapid development of science and technology, artificial intelligence technology is increasingly used in people's daily lives, bringing more convenience to our daily lives.

[0003] In the related art, most smart devices such as in-vehicle systems, voice robots, and mobile phones have the function of recognizing and executing voice commands, allowing users to interact with the smart devices through a voice dialogue system.

[0004] However, in the related art, there is a problem of low accuracy in voice information recognition during voice conversations. Summary of the invention

[0005] Based on this, it is necessary to provide an artificial intelligence entity, a voice information recognition method, a storage medium and a program product that can improve the accuracy of voice information recognition in response to the above technical problems.

[0006] In a first aspect, the present application provides an artificial intelligence entity. The artificial intelligence entity includes a memory and a processor, wherein a computer program is stored in the memory, and the processor is used to implement the following steps when calling and executing the computer program:

[0007] In response to the target voice information, identifying semantic information of the target voice information through a preset semantic recognition model;

[0008] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0009] Response information is determined according to the target voice information and semantic information, and the response information is output.

[0010] In one embodiment, the processor further implements the following steps when invoking and executing the computer program:

[0011] Obtain corpus from various human-computer interaction scenarios;

[0012] Extract corresponding triple information from the corpus in each human-computer interaction scenario; triple information includes entities, relations and attributes;

[0013] Construct a semantic knowledge graph based on triple information.

[0014] In one embodiment, the data expansion process includes: data transformation process and data addition and deletion process;

[0015] The processor also implements the following steps when it is used to call and execute a computer program:

[0016] Obtain small sample labeled data in each human-computer interaction scenario;

[0017] Perform data transformation on small sample labeled data to obtain expanded sample data;

[0018] The expanded sample data is processed by adding and deleting data to obtain large sample labeled data.

[0019] In one embodiment, data transformation processing is performed on the small sample annotated data to obtain expanded sample data, including:

[0020] Determine a target entity and an entity attribute corresponding to the target entity in each small sample annotated data; the target entity includes at least one entity;

[0021] According to the entity attributes corresponding to the target entity, obtain entities of the same type as the target entity from the semantic knowledge graph;

[0022] The target entity in the small sample annotated data is replaced with an entity of the same type to obtain the expanded sample data.

[0023] In one embodiment, replacing the target entity in the small sample annotated data with an entity of the same type includes:

[0024] Obtain the location information of the target entity in the small sample annotation data;

[0025] According to the location information, the target entity in the small sample annotation data is replaced with an entity of the same type.

[0026] In one embodiment, data addition and deletion processing is performed on the expanded sample data to obtain large sample annotated data, including:

[0027] Based on a preset prefix and suffix database, randomly add a prefix and / or a suffix to the expanded sample data to obtain first expanded data;

[0028] Modal particles are added and deleted on the first expanded data to obtain large sample labeled data.

[0029] In one embodiment, determining the response information according to the target voice information and the semantic information includes:

[0030] determining a reference entity and attributes of the reference entity in the target speech information;

[0031] According to the attributes of the reference entity, obtaining at least one candidate entity corresponding to the reference entity from the semantic knowledge graph;

[0032] Determine the entity with the highest similarity to the reference entity among at least one candidate entity as the standard entity;

[0033] The response information is determined according to the semantic information of the standard entity and the target voice information.

[0034] In one embodiment, the processor further implements the following steps when invoking and executing the computer program:

[0035] According to the preset cycle, the semantic knowledge graph is updated based on the corpus in each human-computer interaction scenario at the current moment.

[0036] In a second aspect, the present application further provides a method for voice information recognition, which includes the steps implemented by any artificial intelligence entity in the first aspect.

[0037] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps implemented by any one of the artificial intelligence entities in the first aspect.

[0038] In a fourth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the computer program implements the steps implemented by any one of the artificial intelligence entities in the first aspect.

[0039] The above-mentioned artificial intelligence entity, speech information recognition method, storage medium and program product, in response to the target speech information, recognizes the semantic information of the target speech information through a preset semantic recognition model; determines the response information according to the target speech information and the semantic information, and outputs the response information. Among them, the semantic recognition model is obtained by training with large sample annotated data, and the large sample annotated data is obtained by data expansion processing of the small sample annotated data through the pre-constructed semantic knowledge graph; the semantic knowledge graph includes corpora in a variety of human-computer interaction scenarios. In this application, a semantic knowledge graph is constructed based on corpora in a variety of human-computer interaction scenarios, and the semantic knowledge graph contains a large number of entities and entity relationships in the language text generated in human-computer interaction, so that the constructed semantic knowledge graph can provide data support for the learning of the semantic recognition model. Furthermore, since the small sample annotated data is a language sentence pattern that has been manually annotated, after obtaining a large amount of corpus from the semantic knowledge graph, it can be automatically annotated with reference to the language sentence pattern annotated in the small sample annotated data to obtain a large sample annotated data, which reduces the annotation cost of manually annotating massive data, improves the annotation rate, and also improves the reliability of the annotated data. In addition, by expanding the processed large sample annotated data to train the semantic recognition model, the semantic recognition model can learn more interactive corpus, improving the training effect of the semantic recognition model. In this way, in the human-computer voice dialogue scenario, after the artificial intelligence receives the target voice information, it recognizes and analyzes the target voice information through the trained semantic recognition model, thereby accurately determining the semantic information of the target voice information and improving the recognition accuracy of the semantic information. Furthermore, based on accurate semantic information and target voice information, the artificial intelligence can output response information that is more in line with the interaction expectations, and the human-computer interaction effect is better. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A diagram showing an application environment of a method for voice information recognition in one embodiment;

[0041] Figure 2 A flowchart of a method for voice information recognition in one embodiment;

[0042] Figure 3 A flowchart of a method for constructing a semantic knowledge graph in one embodiment;

[0043] Figure 4 is a partial schematic diagram of a semantic knowledge graph in one embodiment;

[0044] Figure 5 is a flow chart of a data enhancement method in one embodiment;

[0045] Figure 6 A schematic diagram of data transformation processing in one embodiment;

[0046] Figure 7 A schematic diagram of data transformation processing in another embodiment;

[0047] Figure 8 is a flow chart of a data enhancement method in another embodiment;

[0048] Fig. 9 is a flow chart of a method for voice information recognition in another embodiment;

[0049] Fig.10 is a structural block diagram of a speech information recognition device in one embodiment;

[0050] Fig.11 1 is a diagram of the internal structure of an artificial intelligence entity in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] With the development of science and technology, artificial intelligence technology is increasingly used in people's daily lives, bringing more convenience to people's lives. For example, artificial intelligence can interact with users through a dialogue system to respond to users' voice information, output corresponding response information and / or perform corresponding actions.

[0053] Among them, the dialogue system refers to a computer system that can interact with humans in a coherent manner. The artificial intelligence entity with the dialogue system can be, but is not limited to: a voice assistant that frees the user's hands and performs various actions such as booking tickets and ordering meals; an intelligent customer service that replaces manual customer service and conducts voice dialogues with users; and intelligent terminal devices such as smart speakers and smart projectors. In addition, the above-mentioned artificial intelligence entity can also exist as a separate voice interaction robot to interact with humans in voice in actual application scenarios and provide corresponding services, etc. This application does not limit this.

[0054] In a dialogue system, the artificial intelligence agent needs to be able to recognize the user's voice information and the user's semantic information, and then query the response information required by the user from the database based on the entities in the semantic information and voice information, and output the response information to the user.

[0055] In the related art, artificial intelligence can recognize and analyze the user's voice information through semantic recognition models. The recognition effect of the semantic recognition model is based on the quality and quantity of the model training data. Therefore, in order to improve the recognition effect of the semantic recognition model, a large amount of labeled data is needed to train it.

[0056] However, in some fields or human-computer interaction scenarios, the speech information generated in human-computer interaction is relatively small and difficult to obtain, resulting in less data for semantic recognition model training. Furthermore, since the training data needs to be manually annotated in advance, the data annotation cost is high, and in order to minimize human errors in the annotation process, certain professional requirements are placed on the annotators, making annotation more difficult.

[0057] In addition, the semantic recognition model trained with massive sample annotated data still has the problem of limited generalization ability. It cannot have the broad prior knowledge and logical reasoning ability like humans. When performing speech information recognition and analysis, there are semantic recognition errors and / or entity extraction errors. Part of the reason is that there are errors in the annotated data, and the other part is that although the annotated data is massive, it is still limited. With the rapid development of society, new language vocabulary will continue to emerge, resulting in the need for new annotated data to retrain the semantic recognition model.

[0058] Based on this, the present application provides an artificial intelligence entity, a voice information recognition method, a storage medium and a program product, so that when the artificial intelligence entity has a voice conversation with a user, the user's target voice information is recognized and analyzed through a semantic recognition model, thereby accurately determining the semantic information of the target voice information and improving the accuracy of semantic recognition.

[0059] The speech information recognition method provided in the present application can be applied to an artificial intelligence agent. The artificial intelligence agent can be any agent that realizes human-computer voice dialogue in a real environment. The agent can be used as an intelligent component in a terminal, such as an intelligent customer service and voice assistant in a terminal. The agent can also be used as a separate terminal, such as an intelligent speaker and a voice interaction robot.

[0060] As an example, the internal structure of the artificial intelligence entity is as follows Figure 1 As shown, the processor in the internal structure is used to respond to the received voice information and output the response information. The memory in the internal structure includes a non-volatile storage medium and an internal memory, and the non-volatile storage medium stores an operating system, a computer program and a database; the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database is used to store data such as corpora in various human-computer interaction scenarios. The network interface is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the voice information recognition method provided in the present application is implemented.

[0061] The following will use examples and drawings to specifically describe in detail the technical solutions of the embodiments of the present application and how the technical solutions of the embodiments of the present application solve the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. It should be noted that the embodiment of the present application provides a method for voice information recognition, and its execution subject can be an artificial intelligence entity, and can also be a voice information recognition device, which can be implemented as part or all of the processor through software, hardware, or a combination of software and hardware. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments.

[0062] In one embodiment, the present application provides an artificial intelligence entity. The artificial intelligence entity includes a memory and a processor, and a computer program is stored in the memory. Figure 2 As shown, when the processor calls and executes the computer program, the following steps are implemented:

[0063] Step 210: In response to the target voice information, the semantic information of the target voice information is identified by using a preset semantic recognition model.

[0064] Among them, the semantic recognition model is obtained by training with large sample annotated data, and the large sample annotated data is obtained by data expansion processing of small sample annotated data through a pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios.

[0065] It should be noted that the target voice information can be the voice content of any voice of any user received by the artificial intelligence entity, and the semantic information is the meaning of the target voice information, reflecting the user's true conversation intention.

[0066] In a possible implementation, the implementation process of step 210 may be: the artificial intelligence entity receives the target voice information, performs entity recognition on the target voice information through a semantic recognition model, and analyzes the semantic information of the target voice information based on the relationship between the recognized entities.

[0067] As an example, the target voice information is: "How is the weather in Shanghai today?", then the entities in the target voice information are: "today", "Shanghai" and "weather", and then according to the attributes and relationships between the entities, the semantic information of the target voice information is determined to be: asking about the weather.

[0068] In addition, before step 210, the process of obtaining the semantic recognition model can be: obtaining small sample annotated data that has been manually annotated in various human-computer interaction scenarios; performing data expansion processing on the small sample annotated data through the constructed semantic knowledge graph to obtain a large amount of large sample annotated data; training the initial semantic recognition model through the large sample annotated data to obtain a trained semantic recognition model.

[0069] The large sample annotated data includes multiple groups of annotated data, and each group of annotated data includes a speech text and semantic information of the speech text.

[0070] Furthermore, the implementation process of using large sample annotated data to train the initial semantic recognition model can be: using multiple groups of annotated data in the large sample annotated data in turn as input of the initial semantic recognition model, training the initial semantic recognition model until the semantic information output by the initial semantic recognition model meets the preset convergence conditions, determining that the initial semantic recognition model converges, and obtaining the semantic recognition model.

[0071] In addition, in order to verify the learning effect of the model, the large sample labeled data can also include speech text without labeled semantic information. During the training cycle, the learning effect of the initial semantic recognition model is verified by the speech text without labeled semantic information.

[0072] It should be noted that when training the initial semantic recognition model, supervised learning can be used for iterative training, or unsupervised learning can be used for iterative training, and the embodiments of the present application do not limit this.

[0073] As an example, the training process of supervised learning can be specifically as follows: multiple groups of labeled data are used as input of the initial semantic recognition model, and the initial semantic recognition model is iteratively trained until the error between the semantic information recognized by the initial semantic recognition model and the semantic information annotated in each speech text is less than a preset value, then the preset convergence condition is met, and it is determined that the initial semantic recognition model has converged, and a semantic recognition model is obtained.

[0074] Step 220: Determine response information according to the target voice information and semantic information, and output the response information.

[0075] Among them, the response information can be output in the form of voice, text language, or chart. The specific form of outputting the response information can be determined based on actual conditions or user interaction needs, and this application does not impose any restrictions on this.

[0076] In one possible implementation, the artificial intelligence entity determines the user's true intention to output target voice information to the artificial intelligence entity based on semantic information, and then obtains corresponding response content from the database based on the semantic information and the target voice information to determine the final output response information based on the response content.

[0077] As an example, when the target voice message is: How is the weather in Shanghai today, and the semantic information is: Asking about the weather, the artificial intelligence agent determines the corresponding date based on the entity "today" in the target voice message, determines the location based on the entity "Shanghai", and then obtains the weather conditions in Shanghai today from the database based on the semantic information.

[0078] The weather conditions may be the changes throughout the day or the real-time conditions at the current time. Moreover, the weather conditions include but are not limited to: sunny, rainy, snowy, temperature, humidity, wind direction, wind speed, warmth index, ultraviolet intensity, visibility, and the probability of occurrence of special disastrous weather.

[0079] Furthermore, the weather conditions in Shanghai today can be output in the form of voice broadcast, or can be intuitively displayed in the display interface of the artificial intelligence body in the form of charts, and text description information of the weather conditions can also be directly output.

[0080] It should be understood that the above is only an example of a conversation scenario of querying weather conditions through an artificial intelligence entity, and is not intended to limit the specific application scenarios of the artificial intelligence entity executing the above voice information recognition method.

[0081] In an embodiment of the present application, a semantic knowledge graph is constructed based on corpora in a variety of human-computer interaction scenarios. The semantic knowledge graph contains a large number of entities and entity relationships in the language text generated in human-computer interaction, so that the constructed semantic knowledge graph can provide data support for the learning of the semantic recognition model. Furthermore, since the small sample annotated data is a language sentence pattern that has been manually annotated, after obtaining a large amount of corpus from the semantic knowledge graph, it can be automatically annotated with reference to the language sentence pattern annotated in the small sample annotated data to obtain a large sample annotated data, thereby reducing the annotation cost of manually annotating massive data, improving the annotation rate, and also improving the reliability of the annotated data. In addition, by training the semantic recognition model with the expanded and processed large sample annotated data, the semantic recognition model can learn more interactive corpora, thereby improving the training effect of the semantic recognition model. In this way, in the human-computer voice dialogue scenario, after the artificial intelligence entity receives the target voice information, it recognizes and analyzes the target voice information through the trained semantic recognition model, thereby accurately determining the semantic information of the target voice information and improving the recognition accuracy of the semantic information. Furthermore, based on precise semantic information and target voice information, the artificial intelligence agent can output response information that is more in line with interaction expectations, resulting in better human-computer interaction effects.

[0082] Based on the above embodiments, the present application also provides a method for constructing a semantic knowledge graph, by which a semantic knowledge graph including corpora in various human-computer interaction scenarios is constructed to provide more sample annotation data for training semantic recognition models.

[0083] The concept of Knowledge Graph (KG) was formally proposed by Google in 2012, with the original purpose of optimizing its search engine performance. In essence, the knowledge graph is a semantic network that expresses various types of entities and the semantic relationships between them. In other words, the knowledge graph is a directed graph with different types of entities as nodes and various relationships between entities as edges.

[0084] In one embodiment, Figure 3 As shown, the semantic knowledge graph construction method provided in this application is applied to Figure 1 The artificial intelligence agent in the example is used to illustrate, including the following steps:

[0085] Step 310: Acquire corpora in multiple human-computer interaction scenarios.

[0086] The corpus is the voice text output by different users to the artificial intelligence entity when expressing different interaction intentions in different human-computer interaction scenarios. It should be understood that in the same scenario and with the same semantic information, different users' language expressions and / or expression habits will result in different voice texts.

[0087] As an example, in a human-computer interaction scenario where music is played by an artificial intelligence agent, the semantic information is to play a piece of music. The acquired corpus may include: Hello, please play a piece of music by A; please play a random song by Zhou X; please play a piece of music by Zhou X.

[0088] As another example, in a human-computer interaction scenario where an artificial intelligence agent controls movie playback, the semantic information is to play a movie directed by Li X, and the acquired corpus may include: play Inception directed by Li X; play a movie in which Li X participated; play Li X's xx movie.

[0089] In one possible implementation, the implementation process of step 310 can be: obtaining voice data from various human-computer interaction scenarios through a crawler method, and further pre-processing the voice text output by the user to the artificial intelligence entity (for example, screening, cleaning, conversion, sorting and classification, etc.) to obtain corpora in various human-computer interaction scenarios.

[0090] Step 320: extract corresponding triple information from the corpus in each human-computer interaction scenario; wherein the triple information includes entities, relations and attributes.

[0091] In a possible implementation, the implementation process of step 320 may be: extracting information from the corpus in each human-computer interaction scenario to obtain corresponding structured information such as entities, relationships, and entity attributes. Further, various types of knowledge extracted from the real world are expressed as structured language text that can be stored and calculated by computers through knowledge representation to obtain triple information.

[0092] Among them, information extraction includes entity extraction, relationship extraction and attribute extraction. Entity extraction, also known as named entity recognition (NER), is the automatic identification of named entities from speech text. The main methods include rule-based methods, statistical machine learning methods, and information extraction for open domains. However, after entity extraction of speech text, a series of discrete named entities are obtained. In order to obtain semantic information, it is also necessary to analyze the relationship between entities from the speech text, connect the entities through the relationship, and form a network knowledge structure. The main methods of relationship extraction include artificially constructed grammar and semantic rules, statistical machine learning, and open domain-oriented relationship extraction technology. Furthermore, the purpose of attribute extraction is to collect attribute information of specific entities from different information sources. For example, for a public figure, his nickname, birthday, nationality, educational background and other information can be obtained from public information on the Internet. Attribute extraction technology can collect this information from multiple data sources to achieve a complete outline of entity attributes.

[0093] Furthermore, knowledge representation usually uses Resource Description Framework (RDF) triples (Subject-Property-Object, SPO) to symbolically describe the relationship between entities.

[0094] Specifically, a triple can be represented in the form of (entity, attribute, attribute value) or in the form of (entity, relationship, entity), and the embodiments of the present application do not limit this.

[0095] Step 330: Construct a semantic knowledge graph based on the triple information.

[0096] Among them, the constructed semantic knowledge graph can be a knowledge graph that meets a single human-computer interaction scenario, or it can be a comprehensive knowledge graph that meets multiple human-computer interaction scenarios. The embodiments of this application do not impose any restrictions on this.

[0097] In addition, when constructing a semantic knowledge graph, it is necessary to first set the schema of the semantic knowledge graph to limit the format of the corpus to be added to the semantic knowledge graph. In other words, determine the corpus model for each human-computer interaction scenario, which contains the meaningful entity types in the scenario and the entity attributes corresponding to these entity types.

[0098] In this way, by setting a pattern to standardize the structured expression of the corpus in the semantic knowledge graph, a piece of corpus must satisfy the pre-defined entity objects and types in the pattern before it is allowed to be updated into the semantic knowledge graph.

[0099] In addition, it should be noted that by executing the above step 320, knowledge elements such as entities, relationships and attributes can be extracted from the original corpus, but these results may contain a large amount of redundant and erroneous information, and the relationship between the data is also flat, lacking hierarchy and logic.

[0100] Therefore, in a possible implementation, the implementation process of step 330 can be: through knowledge fusion, the entities, relationships and attributes in the above triple information are cleaned and integrated, the ambiguity between entity referents and entity objects is eliminated, and a series of basic fact expressions are obtained. Further, the basic facts after knowledge fusion are processed, and the knowledge processing includes ontology construction, knowledge reasoning, quality assessment, etc., to obtain a structured and networked semantic knowledge graph.

[0101] As an example, Figure 4 As shown, for the music field, the triple information composed of the relevant information of singer A includes: (singer A, identity, singer), (singer A, identity, producer), (singer A, identity, director), (singer A, birthplace, Taiwan), (singer A, ethnicity, Han nationality), (singer A, nationality, China), (singer A, singing, music 1), (singer A, singing, music 2), (singer A, singing, music 3) and (singer A, singing, music 4). Further, for singer A, his wife is actor E, so there is also triple information (actor E, couple, singer A); for music 4, singer B and singer C have also sung, so there is also triple information (singer B, singing, music 4) and (singer C, singing, music 4).

[0102] It should be understood that Figure 4 The semantic knowledge graph corresponding to the triple information in the above example is just a schematic diagram of the local semantic knowledge graph in the human-computer interaction scenario of playing music through an artificial intelligence entity. The semantic knowledge graph can also include other triple information.

[0103] In this embodiment, by obtaining corpora in various human-computer interaction scenarios and extracting entities, relationships and attributes therefrom, a plurality of triple information is obtained. Further, a semantic knowledge graph is constructed based on the triple information. In this way, the constructed semantic knowledge graph contains entities and relationships between entities in speech texts in various human-computer interaction scenarios, which can provide data support for the learning of semantic recognition models.

[0104] In addition, if the semantic recognition model wants to achieve the ideal recognition accuracy, it needs to use a lot of annotated data for learning. Therefore, this application expands a limited number of small sample annotated data to obtain a large amount of large sample annotated data to provide data support for the learning of the semantic recognition model. In one possible implementation method, the small sample annotated data can be expanded through the constructed semantic knowledge graph to obtain large sample annotated data, thereby achieving the purpose of annotated data enhancement. Among them, the expansion processing includes data transformation processing and data addition and deletion processing.

[0105] It should be noted that data augmentation, as a data preprocessing method, is widely used in the field of computer vision. For example, image samples can be rotated, cropped, flipped, and translated. Using data augmentation can effectively improve the generalization ability of the model and reduce the demand for labeled data. However, in the field of natural language processing, there are fewer data augmentation methods, and different data augmentation methods need to be designed for different tasks.

[0106] In one embodiment, Figure 5 As shown, the present application provides a data enhancement method, which is applied to Figure 1 The artificial intelligence agent in the example is used to illustrate, including the following steps:

[0107] Step 510: Obtain small sample annotation data in each human-computer interaction scene.

[0108] Among them, each annotation data in the small sample annotation data includes a speech text and corresponding semantic information.

[0109] It should be noted that in the same interaction scenario, the semantic information in the human-computer interaction process may be the same, but different users have different ways of expression and / or habits of expression. For the same semantics, the voice texts transmitted to the artificial intelligence entity by different users are different.

[0110] The above-mentioned small sample annotation data can be the voice text collected by the artificial intelligence entity in various human-computer interaction scenarios, and the sample annotation data stored in the database after manual annotation, or it can be the labeled sample annotation data directly obtained from other third-party systems or databases. The embodiment of the present application does not limit its source.

[0111] Step 520: Perform data transformation processing on the small sample labeled data to obtain augmented sample data.

[0112] In a possible implementation manner, the implementation process of step 520 can be: determine the target entity and the entity attributes corresponding to the target entity in each small sample labeled data; according to the entity attributes corresponding to the target entity, obtain the entities of the same type as the target entity from the semantic knowledge graph; replace the target entity in the small sample labeled data with the entity of the same type to obtain the augmented sample data.

[0113] Among them, the target entity includes at least one entity, and the entities of the same type include multiple entities. In other words, based on one target entity, multiple entities of the same type can be obtained from the semantic knowledge graph, and the entity attributes of the target entity and the entities of the same type are the same.

[0114] As an example, as Figure 6 shown, the voice text in the obtained labeled data is: Play a piece of work A by Wang XX, and the semantic information is: play music. Then the first target entity in this labeled data is "Wang XX", and the second target entity is "Work A". Moreover, the attribute of the entity "Wang XX" is singer, the attribute of the entity Work A is music, and there is a singing relationship between the first target entity and the second entity.

[0115] Based on the first target entity, the entities of the same type with the attribute of "singer" obtained from the semantic knowledge graph include but are not limited to: Zhou XX, Zhang X, Cai XX, and Lin XX. For each entity of the same type of the first target entity, the entities of the same type with the attribute of "music" obtained from the knowledge graph include but are not limited to: Work B, Work C, Work D, and Work E.

[0116] Furthermore, in a possible implementation manner, the implementation process of replacing the target entity in the small sample labeled data with the entity of the same type can be: obtain the position information of the target entity in the small sample labeled data; according to the position information, replace the target entity in the small sample labeled data with the entity of the same type.

[0117] See Figure 6 , replace "Wang XX" in the voice text with: Zhou XX, Zhang X, Cai XX, and Lin XX in turn, and replace "Work A" with Work B, Work C, Work D, and Work E in turn.

[0118] It should be noted that during the replacement process, the singing relationship existing between the first target entity and the second target entity should be kept unchanged. For example, in the above example, when replacing the first target entity with "Zhou XX", the second target entity can only be replaced with Work B to ensure the accuracy of the generated large sample labeled data after replacement.

[0119] Based on the above example, after data change processing, the expanded sample data obtained from the small sample labeled data "Work A from Wang XX" includes: Work B from Zhou XX; Work D from Cai XX; Work C from Zhang X; and Work E from Lin XX.

[0120] In another possible implementation, the implementation process of replacing the target entity in the small sample annotated data with an entity of the same type can be: according to the semantic information corresponding to the speech text in the small sample annotated data, the small sample annotated data is classified to obtain multiple categories of annotated data, and the semantic information of each category of annotated data is the same. Then, for multiple speech texts with the same semantic information, the target entity and the attributes of the target entity included in each speech text are determined, and the position of the target entity is left blank, and the attributes of the entity that needs to be filled in the position are noted in the blank position, so that multiple sentence templates under the speech text can be obtained. Furthermore, according to the attributes of the fillable entity positions marked in each sentence template, and the entity relationships between the fillable entities, multiple entities of the same type that meet the requirements are obtained from the semantic knowledge graph, and the entities of the same type are filled into each sentence template in turn, so as to obtain the expanded sample data.

[0121] As an example, suppose the semantic information in the small sample annotated data is: play music. The annotated speech texts under this type of semantic information include the following:

[0122] (1) Could you please play Zhou XX’s work B?

[0123] (2) Let’s listen to a piece by C, the one by Zhang X.

[0124] (3) Play any song from Lin XX’s collection Y.

[0125] Among them, the target entities in the (1) speech text are: Zhou XX and B's work, and the attributes are: singer and music respectively. The first sentence template obtained is: Please play [music] by [singer], okay? .

[0126] Similarly, the target entities in the second speech text are: C works and Zhang X, and their attributes are: music and singer, respectively. The second sentence template obtained is: Let’s listen to [music], which is [singer]. The target entities in the third speech text are: Lin XX and Y works collection, and their attributes are: singer and album, respectively. The third speech sentence obtained is: Let’s listen to any song from [singer’s] [album].

[0127] Furthermore, if Figure 7As shown, according to the entity attributes of the target entity in the first sentence pattern template and the relationships between the target entities, the first type of entities with the attribute of "singer" obtained from the semantic knowledge graph can be: Wang XX, Zhang X, Cai XX, and Lin XX. Then, according to the singing relationship between "singer" and "music", obtain the songs sung by Wang XX, Zhang X, Lin XX, and Cai XX, and the second type of entities with the attribute of "music" obtained can be: Work A, Work C, Work D, and Work E. Finally, fill the above first type of entities into the positions marked with the "singer" attribute in the first sentence pattern template in sequence, and fill the second type of entities into the positions marked with the "music" attribute in the first sentence pattern template in sequence, then the augmented sample data can be obtained.

[0128] In this way, the augmented sample data obtained through the first sentence pattern template includes: Could you please play Work C by Zhang X?, Could you please play Work E by Lin XX?, Could you please play Work D by Cai XX?, Could you please play Work A by Wang XX?.

[0129] Step 530: Perform data addition and deletion processing on the augmented sample data to obtain large sample labeled data.

[0130] Among them, for the augmented sample data, the addition and deletion processing includes adding text content to the speech text and deleting the text content of the speech text.

[0131] In a possible implementation manner, the implementation process of step 530 can be: Based on a preset prefix and suffix database, randomly add prefixes and / or suffixes to the augmented sample data to obtain the first augmented data; perform the operation of adding and deleting modal particles on the first augmented data to obtain the large sample labeled data.

[0132] Among them, the prefix and suffix database includes a prefix library and a suffix library. The prefix library includes multiple words that can be added to the beginning of the speech text, such as: handsome guy, beautiful lady, hello, hello there, are you there. The suffix library includes multiple words that can be added to the end of the speech text, such as: okay, can you, can't you, okay?, please, please.

[0133] As an example, the modal particles in the speech text include: um, of, already,呢,吧, ah, oh, etc. meaningless words that do not affect the content but may involve personal expression habits and ways.

[0134] In the embodiments of the present application, based on the constructed semantic knowledge graph, through data change processing and data addition and deletion processing on the small sample labeled data, the large sample labeled data is obtained. In this way, through the already labeled small sample labeled data and the semantic knowledge graph, a large amount of large sample labeled data can be automatically obtained, reducing the manual labeling cost while also improving the data labeling speed and accuracy.

[0135] based on Figure 5 The embodiment shown, as Figure 8 As shown, the present application also provides another data enhancement method, which is applied to Figure 1 The artificial intelligence agent in the example is used to illustrate, including the following steps:

[0136] Step 810: Obtain small sample annotation data in each human-computer interaction scene;

[0137] Step 820: Determine the target entity and the entity attribute corresponding to the target entity in each small sample annotated data; the target entity includes at least one entity;

[0138] Step 830: According to the entity attributes corresponding to the target entity, obtain entities of the same type as the target entity from the semantic knowledge graph;

[0139] Step 840: Obtaining location information of the target entity in the small sample annotated data;

[0140] Step 850: According to the position information, the target entity in the small sample annotated data is replaced with an entity of the same type to obtain expanded sample data;

[0141] Step 860: Based on a preset prefix and suffix database, randomly add a prefix and / or a suffix to the extended sample data to obtain first extended data;

[0142] Step 870: Add and delete modal particles on the first expanded data to obtain large sample labeled data.

[0143] The implementation principle and technical effects of each step in the data enhancement method provided in this embodiment are similar to those in the previous embodiment. Figure 5 The method embodiments shown are similar and will not be described again here.

[0144] Based on the data enhancement method shown in the above embodiment, the small sample annotated data is expanded to obtain a large amount of large sample annotated data. Further, the large sample annotated data is used to train the initial semantic recognition model to obtain the semantic recognition model of the present application. In this way, after obtaining the semantic recognition model, the artificial intelligence entity can use the semantic recognition model to accurately recognize and respond to the target voice information.

[0145] In one embodiment, Fig. 9 As shown, the implementation process of determining the response information according to the target voice information and semantic information in the above step 220 includes the following steps:

[0146] Step 910: Determine the reference entity and the attributes of the reference entity in the target voice information.

[0147] The reference entity is at least one entity in the target voice information, and the attribute is a word that describes the characteristics of the entity from different dimensions.

[0148] As an example, if the target voice information is: play Jay Chou's "Eviction" for listening, the artificial intelligence entity will identify the reference entities including: Jay Chou and "Eviction" based on the semantic recognition model, and the attributes are "singer" and "music".

[0149] Step 920: According to the attributes of the reference entity, obtain at least one candidate entity corresponding to the reference entity from the semantic knowledge graph.

[0150] It should be noted that the candidate entity obtained from the semantic knowledge graph may or may not be the same as the reference entity. This embodiment is intended to illustrate that the attributes of the reference entity and the candidate entity are the same, and does not limit their content text to be the same.

[0151] Step 930: Determine the entity with the highest similarity to the reference entity among at least one candidate entity as the standard entity.

[0152] In this step, if the candidate entity is the same as the reference entity, the reference entity is used as the standard entity; if the candidate entity is different from the reference entity, the candidate entity with the highest similarity is determined as the standard entity based on the similarity between the candidate entity and the reference entity.

[0153] That is, the embodiment of the present application takes the semantic knowledge graph as the standard, and corrects the recognition results of the semantic recognition model in the artificial intelligence entity through the candidate entities in the semantic knowledge graph. When the two are different, the candidate entity with the highest similarity in the semantic knowledge graph is used as the standard entity, and the reference entity recognized by the semantic recognition model is ignored.

[0154] Based on the above example, the target voice information is: play Zhou XX's yyy to listen, the reference entity is "yyy to", and the attribute is "music". However, when searching for music sung by Zhou XX in the semantic knowledge graph, the candidate entity obtained is "yyy", and finally "yyy" is determined as the standard entity.

[0155] Step 940: Determine response information based on the semantic information of the standard entity and the target voice information.

[0156] Based on the above example, the target voice information is: play Zhou XX's yyy for you to listen to, and the semantic information is to play music, then the response information determined by the artificial intelligence entity may be: OK, I will play Zhou XX's song yyy for you.

[0157] In this embodiment, the artificial intelligence entity recognizes the target voice information through a semantic recognition model to determine the reference entity, the attributes of the reference entity, and the semantic information of the target voice information. Then, based on the entity attributes of the reference entity, at least one candidate entity is obtained from the semantic knowledge graph, and the candidate entity is used to correct the reference entity recognized by the artificial intelligence entity to determine the standard entity from the reference entity and the candidate entity. Furthermore, based on the standard entity and the semantic information of the target voice information, the response information that better meets the user's interaction expectations is determined, so that the human-computer interaction effect is better.

[0158] Based on the above embodiments, since this application needs to use the semantic knowledge graph to correct the recognition results of the semantic recognition model to improve the recognition accuracy of the semantic recognition model. Therefore, with the emergence of new corpus, this application also needs to update the constructed semantic knowledge graph to provide more reference information for the semantic recognition model.

[0159] In one embodiment, the implementation process of updating the semantic knowledge graph may be: updating the semantic knowledge graph according to the corpus in each human-computer interaction scenario at the current moment according to a preset period.

[0160] In one possible implementation, the corpus in each human-computer interaction scenario is obtained regularly, and the semantic knowledge graph is fully or incrementally updated based on the triple information in the corpus at the current moment, so that the semantic knowledge graph can be supplemented with more entities and the relationships between entities can be enriched on the original basis.

[0161] Optionally, after updating the semantic knowledge graph, the updated semantic knowledge graph can be used to further expand the original large annotated data, and the expanded annotated data can be used to retrain the semantic recognition model.

[0162] In the embodiment of the present application, by updating the semantic knowledge graph, on the one hand, more training samples can be provided for the semantic recognition model. On the other hand, the entities in the updated semantic knowledge graph are used to correct the entity recognition results of the semantic recognition model. By updating the semantic knowledge graph, the recognition accuracy of the semantic recognition model can be guaranteed while reducing the number of semantic recognition model training times.

[0163] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0164] In addition, an embodiment of the present application also provides a voice information recognition method. The implementation solution to the problem provided by the method is similar to the implementation process of voice information recognition by the above-mentioned artificial intelligence entity. Therefore, the specific limitations in one or more voice information recognition method embodiments provided below can be referred to the limitations of the implementation steps when the processor in the artificial intelligence entity calls and executes computer-readable instructions above, and will not be repeated here.

[0165] It should be noted that the voice information recognition method provided in this application can be executed by an artificial intelligence entity or by other computer devices. This application does not limit the execution subject. Any computer device that performs voice recognition according to the voice information recognition method shown in this application, or implements human-computer voice dialogue interaction, is within the protection scope of this application.

[0166] The speech information recognition method provided in this application comprises the following steps:

[0167] In response to the target voice information, identifying semantic information of the target voice information through a preset semantic recognition model;

[0168] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0169] Response information is determined according to the target voice information and semantic information, and the response information is output.

[0170] In one embodiment, the method further comprises:

[0171] Obtain corpus from various human-computer interaction scenarios;

[0172] Extract corresponding triple information from the corpus in each human-computer interaction scenario; triple information includes entities, relations and attributes;

[0173] Construct a semantic knowledge graph based on triple information.

[0174] In one embodiment, the data expansion process includes: data transformation process and data addition and deletion process; the method further includes:

[0175] Obtain small sample labeled data in each human-computer interaction scenario;

[0176] Perform data transformation on small sample labeled data to obtain expanded sample data;

[0177] The expanded sample data is processed by adding and deleting data to obtain large sample labeled data.

[0178] In one embodiment, data transformation processing is performed on the small sample annotated data to obtain the expanded sample data, including:

[0179] Determine a target entity and an entity attribute corresponding to the target entity in each small sample annotated data; the target entity includes at least one entity;

[0180] According to the entity attributes corresponding to the target entity, obtain entities of the same type as the target entity from the semantic knowledge graph;

[0181] The target entity in the small sample annotated data is replaced with an entity of the same type to obtain the expanded sample data.

[0182] In one embodiment, replacing the target entity in the small sample annotated data with an entity of the same type includes:

[0183] Obtain the location information of the target entity in the small sample annotation data;

[0184] According to the location information, the target entity in the small sample annotation data is replaced with an entity of the same type.

[0185] In one embodiment, data addition and deletion processing is performed on the expanded sample data to obtain large sample annotated data, including:

[0186] Based on a preset prefix and suffix database, randomly add a prefix and / or a suffix to the expanded sample data to obtain first expanded data;

[0187] Modal particles are added and deleted on the first expanded data to obtain large sample labeled data.

[0188] In one embodiment, determining the response information according to the target voice information and the semantic information includes:

[0189] determining a reference entity and attributes of the reference entity in the target speech information;

[0190] According to the attributes of the reference entity, obtaining at least one candidate entity corresponding to the reference entity from the semantic knowledge graph;

[0191] Determine the entity with the highest similarity to the reference entity among at least one candidate entity as the standard entity;

[0192] The response information is determined according to the semantic information of the standard entity and the target voice information.

[0193] In one embodiment, the method further comprises:

[0194] According to the preset cycle, the semantic knowledge graph is updated based on the corpus in each human-computer interaction scenario at the current moment.

[0195] The implementation principle and technical effects of the voice information recognition method provided in this embodiment are similar to the steps performed by the above-mentioned artificial intelligence entity, and will not be repeated here.

[0196] Based on the same inventive concept, the embodiment of the present application also provides a voice information recognition device for implementing the voice information recognition method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more voice information recognition device embodiments provided below can refer to the limitations of the voice information recognition method above, and will not be repeated here.

[0197] In one embodiment, Fig.10 As shown, a speech information recognition device is provided, the device 1000 includes: a semantic recognition module 1010 and a response module 1020, wherein:

[0198] The semantic recognition module 1010 is used to recognize the semantic information of the target voice information through a preset semantic recognition model in response to the target voice information;

[0199] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0200] The response module 1020 is used to determine response information according to the target voice information and semantic information, and output the response information.

[0201] In one embodiment, the apparatus 1000 further includes:

[0202] The first acquisition module is used to acquire corpora in various human-computer interaction scenarios;

[0203] The extraction module is used to extract corresponding triple information from the corpus in each human-computer interaction scenario; the triple information includes entities, relations and attributes;

[0204] A building module is used to construct a semantic knowledge graph based on triple information.

[0205] In one embodiment, the data expansion process includes: data transformation process and data addition and deletion process; the device 1000 also includes:

[0206] The second acquisition module is used to obtain small sample annotation data in each human-computer interaction scene;

[0207] The data transformation module is used to perform data transformation processing on the small sample annotation data to obtain the expanded sample data;

[0208] The data addition and deletion module is used to perform data addition and deletion processing on the expanded sample data to obtain large sample labeled data.

[0209] In one embodiment, the data transformation module includes:

[0210] A first determination unit is used to determine a target entity and an entity attribute corresponding to the target entity in each small sample annotated data; the target entity includes at least one entity;

[0211] A first acquisition unit is used to acquire entities of the same type as the target entity from the semantic knowledge graph according to entity attributes corresponding to the target entity;

[0212] The replacement unit is used to replace the target entity in the small sample annotation data with an entity of the same type to obtain the expanded sample data.

[0213] In one embodiment, the replacement unit comprises:

[0214] An acquisition subunit is used to obtain the location information of the target entity in the small sample annotation data;

[0215] The replacement subunit is used to replace the target entity in the small sample annotation data with an entity of the same type according to the position information.

[0216] In one embodiment, the data adding and deleting module includes:

[0217] An adding unit, configured to randomly add a prefix and / or a suffix to the expanded sample data based on a preset prefix and suffix database to obtain first expanded data;

[0218] The adding and deleting unit is used to perform adding and deleting operations on the first expanded data to obtain large sample labeled data.

[0219] In one embodiment, the response module 1020 includes:

[0220] A second determining unit, used to determine a reference entity and an attribute of the reference entity in the target speech information;

[0221] A second acquisition unit, configured to acquire at least one candidate entity corresponding to the reference entity from the semantic knowledge graph according to the attribute of the reference entity;

[0222] A third determining unit, configured to determine an entity among at least one candidate entity that has the highest similarity to the reference entity as a standard entity;

[0223] The fourth determining unit is used to determine the response information according to the semantic information of the standard entity and the target voice information.

[0224] In one embodiment, the device 1000 further includes:

[0225] The updating module is used to update the semantic knowledge graph according to the preset period and the corpus in each human-computer interaction scenario at the current moment.

[0226] Each module in the above-mentioned speech information recognition device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0227] In one embodiment, an artificial intelligence entity is provided. The artificial intelligence entity can be any intelligent entity or intelligent terminal. Its internal structure diagram can be as follows: Fig.11 As shown. The artificial intelligence entity includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the artificial intelligence entity is used to provide computing and control capabilities. The memory of the artificial intelligence entity includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the artificial intelligence entity is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for voice information recognition is implemented. The display screen of the artificial intelligence entity can be a liquid crystal display screen or an electronic ink display screen, and the input device of the artificial intelligence entity can be a touch layer covering the display screen, or a key, trackball or touchpad set on the shell of the artificial intelligence entity, or an external keyboard, touchpad or mouse, etc.

[0228] Those skilled in the art will understand that Fig.11The structure shown in the figure is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the artificial intelligence entity to which the solution of the present application is applied. The specific artificial intelligence entity may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0229] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0230] In response to the target voice information, identifying semantic information of the target voice information through a preset semantic recognition model;

[0231] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0232] Response information is determined according to the target voice information and semantic information, and the response information is output.

[0233] The computer device provided in the above embodiment has an implementation principle and technical effects similar to those of the above method embodiment, which will not be described in detail here.

[0234] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0235] In response to the target voice information, identifying semantic information of the target voice information through a preset semantic recognition model;

[0236] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0237] Response information is determined according to the target voice information and semantic information, and the response information is output.

[0238] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0239] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0240] In response to the target voice information, identifying semantic information of the target voice information through a preset semantic recognition model;

[0241] The semantic recognition model is obtained by training with large sample annotated data, which is obtained by augmenting the small sample annotated data with the pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in various human-computer interaction scenarios;

[0242] Response information is determined according to the target voice information and semantic information, and the response information is output.

[0243] The above embodiment provides a computer program product, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0244] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0245] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0246] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. An artificial intelligence entity, characterized in that: The artificial intelligence entity includes a memory and a processor, wherein the memory stores a computer program, and the processor is used to implement the following steps when calling and executing the computer program: In response to the target voice information, identifying semantic information of the target voice information by using a preset semantic recognition model; The semantic recognition model is obtained by training with large sample annotated data, and the large sample annotated data is obtained by performing data expansion processing on the small sample annotated data through a pre-built semantic knowledge graph; the semantic knowledge graph includes corpora in a variety of human-computer interaction scenarios; Determining response information according to the target voice information and the semantic information, and outputting the response information; The method of determining the response information based on the target voice information and the semantic information includes: determining a reference entity in the target voice information and attributes of the reference entity; obtaining at least one candidate entity corresponding to the reference entity from the semantic knowledge graph based on the attributes of the reference entity; determining the entity with the highest similarity to the reference entity among the at least one candidate entity as the standard entity; and determining the response information based on the standard entity and the semantic information of the target voice information.

2. The artificial intelligence entity according to claim 1, characterized in that: The processor further implements the following steps when being used to call and execute the computer program: Obtain corpus from various human-computer interaction scenarios; Extracting corresponding triple information from the corpus in each of the human-computer interaction scenarios; the triple information includes entities, relations and attributes; The semantic knowledge graph is constructed according to the triple information.

3. The artificial intelligence entity according to claim 1 or 2, characterized in that: The data expansion process includes: data transformation process and data addition and deletion process; The processor further implements the following steps when being used to call and execute the computer program: Obtaining small sample labeled data in each of the human-computer interaction scenarios; Performing the data transformation process on the small sample labeled data to obtain expanded sample data; The data addition and deletion processing is performed on the expanded sample data to obtain the large sample labeled data.

4. The artificial intelligence entity according to claim 3, characterized in that: The performing the data transformation process on the small sample annotated data to obtain the expanded sample data includes: Determine a target entity in each of the small sample annotated data and an entity attribute corresponding to the target entity; the target entity includes at least one entity; According to the entity attributes corresponding to the target entity, obtaining entities of the same type as the target entity from the semantic knowledge graph; The target entity in the small sample annotated data is replaced with the entity of the same type to obtain the expanded sample data.

5. The artificial intelligence entity according to claim 4, characterized in that: The step of replacing the target entity in the small sample annotated data with the entity of the same type includes: Obtaining location information of the target entity in the small sample annotated data; According to the position information, the target entity in the small sample annotated data is replaced with the entity of the same type.

6. The artificial intelligence entity according to claim 3, characterized in that: Performing the data addition and deletion processing on the expanded sample data to obtain the large sample labeled data includes: Based on a preset prefix and suffix database, randomly add a prefix and / or a suffix to the extended sample data to obtain first extended data; Modal particles are added and deleted on the first expanded data to obtain the large sample labeled data.

7. The artificial intelligence entity according to claim 1 or 2, characterized in that: The processor further implements the following steps when being used to call and execute the computer program: According to a preset cycle, the semantic knowledge graph is updated based on the corpus in each of the human-computer interaction scenarios at the current moment.

8. A method for voice information recognition, characterized in that: The method comprises: the steps implemented by the artificial intelligence agent described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is called and executed by a processor, the steps implemented by the artificial intelligence agent in any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the computer program is called and executed by a processor, the steps implemented by the artificial intelligence agent in any one of claims 1 to 7 are implemented.

Images