Granted Patent B1

US 12,386,914 · App. 19/073,842 · Granted Aug 12, 2025

Inventors: Lucheng Chen (Qingdao, CN); Xiaoping Lu (Qingdao, CN); Zongkui Yan (Qingdao, CN); Yong Wang (Qingdao, CN); Zhenfa Yang (Qingdao, CN); Jian Yang (Qingdao, CN); Mizhen Wang (Qingdao, CN); Pengjing Wang (Qingdao, CN)

Assignees: COSMO INSTITUTE OF INDUSTRIAL INTELLIGENCE (QINGDAO) CO., LTD.; COSMOPlat IoT Technology, Co., Ltd

G06F16/958G06F16/951G06F40/30

Abstract

Claims (88)

1. A construction method of an intelligent interaction service system, applied to an intelligent interaction engine platform, comprising:

obtaining historical query information of a plurality of websites;

obtaining website information of the plurality of websites and an intent identification corresponding to each historical query information;

determining intent category and website association information corresponding to each historical query information according to the historical query information of the plurality of websites and the intent identification corresponding to each historical query information;

inputting historical query information with the same complexity level of website information and the intent category and website association information corresponding to each historical query information into an intent recognition model corresponding to the complexity level of website information for training, to obtain a plurality of trained intent recognition models, and using the plurality of trained intent recognition models as an intent recognition component, wherein different complexity levels of website information correspond to different intent recognition models;

performing an index extraction on the plurality of historical query information according to an index technology to obtain historical query information with an index structure;

establishing an association relationship between the historical query information and the historical query information with the index structure; and

performing an information retrieval training on the association relationship between the historical query information and the historical query information with the index structure according to a retrieval algorithm, to construct an information retrieval component;

obtaining the plurality of website addresses and an access permission of each website address, and crawling website data corresponding to each website address;

filtering first type of website data and second type of website data out from the website data corresponding to each website address, respectively;

determining a website database corresponding to each website address according to the first type of website data corresponding to all website addresses and the second type of website data corresponding to each website address; and

performing a feature extraction on website data corresponding to each website database, respectively, to construct vector database corresponding to each website address, and determining a vector database group according to a plurality of vector databases corresponding to the plurality of website addresses;

constructing an intelligent interaction service system according to a user interaction interface, the intent recognition component, the information retrieval component, the vector database group and a generating model, wherein the generating model is used to generate a natural language response.

2. A website intelligent interaction method, used in the intelligent interaction service system obtained though the construction method according to claim 1 , comprising:

obtaining a user identity, and determining historical multimodal interaction data and a user type that are corresponding to the user identity;

obtaining first modal interaction data sent by a user, and determining a user interaction mode according to the historical multimodal interaction data and the first modal interaction data, wherein the first modal interaction data is used to indicate first interaction data in current interaction data;

performing an intent recognition analysis on current interaction data according to an intent recognition component to obtain a user intent label, and transforming the user intent label into a user intent vector; and

performing a retrieve and match on the user intent vector based on a vector database group to obtain user interaction result information, and sending the user interaction result information corresponding to the user type to a human-computer interaction interface in the user interaction mode.

3. The method according to claim 2 , wherein the determining the user interaction mode according to the historical multimodal interaction data and the first modal interaction data comprises:

performing a feature analysis on the historical multimodal interaction data and the first modal interaction data to obtain a user interaction feature, and determining the user interaction mode according to the user interaction feature.

4. The method according to claim 3 , wherein the historical multimodal interaction data comprises at least one of text data, image data, and audio data, and the performing the feature analysis on the historical multimodal interaction data and the first modal interaction data to obtain the user interaction feature, and determining the user interaction mode according to the user interaction feature comprises:

if the first modal interaction data is the text data, determining, according to the user interaction feature, that the user interaction mode is a search box interaction mode containing a feature identification; and

if the first modal interaction data is the image data or the audio data, matching a digital human role according to the user interaction feature, and determining that the user interaction mode is a digital human role interaction mode.

5. The method according to claim 4 , wherein the matching the digital human role according to the user interaction feature comprises:

obtaining digital human role labels corresponding to a plurality of types of digital humans;

determining a digital human role label with a highest correlation to the user interaction feature according to a correlation between the user interaction feature and the digital human role labels; and

determining that the user interacts with a digital human corresponding to the digital human role label with a highest correlation to the user interaction feature.

6. The method according to claim 2 , wherein the performing the intent recognition analysis on the current interaction data according to the intent recognition component to obtain the user intent label comprises:

obtaining a complexity level of website information corresponding to the current interaction data;

determining a target intent recognition model according to the complexity level of website information corresponding to the current interaction data; and

performing the intent recognition analysis on the current interaction data according to the target intent recognition model to obtain the user intent label.

7. The method according to claim 6 , wherein the determining the target intent recognition model according to the complexity level of website information corresponding to the current interaction data comprises:

if the complexity level of website information corresponding to the current interaction data is a simple task, determining that the target intent recognition model is a machine learning model; and

if the complexity level of website information corresponding to the current interaction data is a complex task, determining that the target intent recognition model is a deep learning model.

8. The method according to claim 2 , wherein the user interaction result information comprises: retrieval vector information and a response content, and the performing the retrieve and match on the user intent vector based on the vector database group to obtain the user interaction result information, and sending the user interaction result information corresponding to the user type to the human-computer interaction interface in the user interaction mode comprises:

performing a correlation search in a target vector database through an information retrieval component according to the user intent vector to obtain a target intent vector corresponding to the user intent vector in the target vector database;

determining the retrieval vector information and a target intent related link, which are corresponding to the target intent vector, according to the target vector database, wherein the target intent related link is used to indicate an information solution link of a plurality of websites associated with an target intent;

inputting the retrieval vector information and the user intent vector into a generating model to obtain the response content; and

sending the response content and the target intent related link, which are corresponding to the user type, to the human-computer interaction interface in the user interaction mode.

9. The method according to claim 8 , wherein the inputting the retrieval vector information and the user intent vector into the generating model to obtain the response content comprises:

performing a structural transformation on the retrieval vector information and the user intent vector to obtain retrieval information and a user intent; and

performing an association analysis on the retrieval information with the user intent to obtain the response content.

10. A website intelligent interaction apparatus, used in the intelligent interaction service system obtained though the construction method according to claim 1 , comprising:

a memory; and

a processor;

wherein the memory stores computer executable instructions;

the processor executes the computer executable instructions stored in the memory to:

obtain a user identity;

determine historical multimodal interaction data and a user type that are corresponding to the user identity;

obtain first modal interaction data sent by a user;

determine a user interaction mode according to the historical multimodal interaction data and the first modal interaction data, wherein the first modal interaction data is used to indicate first interaction data in current interaction data;

perform an intent recognition analysis on current interaction data according to an intent recognition component to obtain a user intent label, and transform the user intent label into a user intent vector;

perform a retrieve and match on the user intent vector based on a vector database group to obtain user interaction result information; and

send the user interaction result information corresponding to the user type to a human-computer interaction interface in the user interaction mode.

11. The apparatus according to claim 10 , wherein the processor is specifically configured to:

perform a feature analysis on the historical multimodal interaction data and the first modal interaction data to obtain a user interaction feature, and determine the user interaction mode according to the user interaction feature.

12. The apparatus according to claim 11 , wherein the historical multimodal interaction data comprises at least one of text data, image data, and audio data, and the performing the feature analysis on the historical multimodal interaction data and the first modal interaction data to obtain the user interaction feature, and the processor is specifically configured to:

if the first modal interaction data is the text data, determine, according to the user interaction feature, that the user interaction mode is a search box interaction mode containing a feature identification; and

if the first modal interaction data is the image data or the audio data, match a digital human role according to the user interaction feature, and determine that the user interaction mode is a digital human role interaction mode.

13. The apparatus according to claim 12 , wherein the processor is specifically configured to:

obtain digital human role labels corresponding to a plurality of types of digital humans;

determine a digital human role label with a highest correlation to the user interaction feature according to a correlation between the user interaction feature and the digital human role labels; and

determine that the user interacts with a digital human corresponding to the digital human role label with a highest correlation to the user interaction feature.

14. The apparatus according to claim 10 , wherein the processor is specifically configured to:

obtain a complexity level of website information corresponding to the current interaction data;

determine a target intent recognition model according to the complexity level of website information corresponding to the current interaction data; and

perform the intent recognition analysis on the current interaction data according to the target intent recognition model to obtain the user intent label.

15. The apparatus according to claim 14 , wherein the processor is specifically configured to:

if the complexity level of website information corresponding to the current interaction data is a simple task, determine that the target intent recognition model is a machine learning model; and

if the complexity level of website information corresponding to the current interaction data is a complex task, determine that the target intent recognition model is a deep learning model.

16. The apparatus according to claim 10 , wherein the user interaction result information comprises: retrieval vector information and a response content, and the processor is specifically configured to:

perform a correlation search in a target vector database through an information retrieval component according to the user intent vector to obtain a target intent vector corresponding to the user intent vector in the target vector database;

determine the retrieval vector information and a target intent related link, which are corresponding to the target intent vector, according to the target vector database, wherein the target intent related link is used to indicate an information solution link of a plurality of websites associated with an target intent;

input the retrieval vector information and the user intent vector into a generating model to obtain the response content; and

send the response content and the target intent related link, which are corresponding to the user type, to the human-computer interaction interface in the user interaction mode.

17. The apparatus according to claim 16 , wherein the processor is specifically configured to:

perform a structural transformation on the retrieval vector information and the user intent vector to obtain retrieval information and a user intent; and

perform an association analysis on the retrieval information with the user intent to obtain the response content.

18. A non-transitory computer storage medium, storing computer executable instructions which, when executed by a processor, implement the following steps:

obtaining a user identity, and determining historical multimodal interaction data and a user type that are corresponding to the user identity;

obtaining first modal interaction data sent by a user, and determining a user interaction mode according to the historical multimodal interaction data and the first modal interaction data, wherein the first modal interaction data is used to indicate first interaction data in current interaction data;

performing an intent recognition analysis on current interaction data according to an intent recognition component to obtain a user intent label, and transforming the user intent label into a user intent vector; and

performing a retrieve and match on the user intent vector based on a vector database group to obtain user interaction result information, and sending the user interaction result information corresponding to the user type to a human-computer interaction interface in the user interaction mode.

19. The non-transitory computer storage medium according to claim 18 , wherein when executed by the processor, the computer executable instructions specifically implement the following steps:

performing a feature analysis on the historical multimodal interaction data and the first modal interaction data to obtain a user interaction feature, and determining the user interaction mode according to the user interaction feature.

20. The non-transitory computer storage medium according to claim 19 , wherein the historical multimodal interaction data comprises at least one of text data, image data, and audio data, and when executed by the processor, and the computer executable instructions specifically implement the following steps:

if the first modal interaction data is the text data, determining, according to the user interaction feature, that the user interaction mode is a search box interaction mode containing a feature identification; and

if the first modal interaction data is the image data or the audio data, matching a digital human role according to the user interaction feature, and determining that the user interaction mode is a digital human role interaction mode.

Assignments (1)

Priority Claims (1)

References Cited (31)