Automatic question-answering method and system

Abstract

The embodiment of the invention provides an automatic question-answering method and system. The method comprises the following steps: generating answer data for a user question from product and service data, aggregating a corresponding data for each product and service data, and setting a weight vector for each path to obtain a corresponding relation table of the path and the weight; when receiving the user question issued by a user, encoding the user question into a feature vector, and then matching the feature vector with various weight vectors in the corresponding relation table of the pathand the weight; extracting an answer of the user question from the answer data of the weight vector corresponding path to feed back to the user when the feature vector is successfully matched with one of the weight vectors. The path and weight vector setting is performed based on the product and service data containing all service data and all user service data but not based on the standard problem base, so that the user question can be well covered, more-comprehensive automatic question-answering can be provided for the user, and the user experience satisfication degree is promoted.

Description

technical field

[0001] The invention relates to the field of artificial intelligence, in particular to an automatic question answering method and system. Background technique

[0002] At present, automatic question answering systems are divided into open domain and closed domain according to the source of their knowledge. The former has a wide range of knowledge sources, including the Internet, domain knowledge documents, etc., and a typical system such as IBM's Watson. The latter knowledge source only includes a specific field, such as customer service robots in the telecommunications field, which only answer knowledge related to the company or industry products.

[0003] At present, in the closed-domain automatic question answering system, the traditional automatic question answering system implementation method is based on standard questions. It extracts the questions that users may ask through manual methods to generate a standard question library StdQ={Qesi, Ansi}, whe...

Examples

Embodiment 1

[0022] The product and service data of each operator is stored in the database. For example, product data such as the user's call balance, call bill, traffic balance, traffic bill, and remaining minutes are stored in the database for each user. For another example, data such as the package content and handling process of telecommunications trains, telecommunications building blocks, telecommunications airplanes, and telecommunications tourism services are also stored in the data. Therefore, the essence of user query is to use natural language to search for the corresponding part of the data content in the database. Taking the telecommunications field as an example, user queries are generally divided into two categories: one is product and service data related to the user itself, such as "how much is my traffic left", "how much is my phone bill left", "my How many call minutes are left in my phone”, “How do I use my phone bill this month”; the other type is to inquire about th...

Embodiment 2

[0049] This embodiment provides an automatic question answering system, please refer to image 3 As shown, it includes a question interaction module 31, a question analysis module 32, and an answer processing module 33;

[0050] The question interaction module 31 is used to obtain the user question issued by the user;

[0051] The problem analysis module 32 is used for encoding the user's problem into a feature vector;

[0052] The answer processing module 33 is used to extract the answer to the user's question from the answer data of the path corresponding to the weight vector when one of the weight vectors in the feature vector and the path and weight correspondence table matches successfully The user; a weight vector corresponds to a path, and a path corresponds to a type of answer data; the answer data is data extracted from product and service data.

[0053] Specifically, the answer processing module 33 includes an answer mapping submodule 331 and an answer processing s...

example 1

[0076] Example 1, automatic question answering process:

[0077] Input: Question from Li Si (13012345002): "Check the phone bill."

[0078] Step 1, the question interaction module initializes the question, and the processing result is: "Check the phone bill."

[0079] Step 2: The processing result of the problem analysis module is: the feature vector of the problem is .

[0080]Step 3: The answer mapping module uses the feature vector of the question as to calculate the cosine similarity with each path vector. The preset similarity threshold is 0.95. The calculation results of each path are as follows:

[0081] User.Call balance: :

[0082]

[0083] User. Call bill details: :

[0084]

[0085] User.traffic balance:

[0086]

[0087] User.Traffic Details :

[0088]

[0089] Telecom building blocks. Flags :

[0090]

[0091] Telecom building blocks. Package content :

[0092]

[0093] Telecom building blocks. Handling process :

[0094]

[0095] Tel...

technical field

[0001] The invention relates to the field of artificial intelligence, in particular to an automatic question answering method and system. Background technique

[0002] At present, automatic question answering systems are divided into open domain and closed domain according to the source of their knowledge. The former has a wide range of knowledge sources, including the Internet, domain knowledge documents, etc., and a typical system such as IBM's Watson. The latter knowledge source only includes a specific field, such as customer service robots in the telecommunications field, which only answer knowledge related to the company or industry products.

[0003] At present, in the closed-domain automatic question answering system, the traditional automatic question answering system implementation method is based on standard questions. It extracts the questions that users may ask through manual methods to generate a standard question library StdQ={Qesi, Ansi}, whe...

Examples

Embodiment 1

[0022] The product and service data of each operator is stored in the database. For example, product data such as the user's call balance, call bill, traffic balance, traffic bill, and remaining minutes are stored in the database for each user. For another example, data such as the package content and handling process of telecommunications trains, telecommunications building blocks, telecommunications airplanes, and telecommunications tourism services are also stored in the data. Therefore, the essence of user query is to use natural language to search for the corresponding part of the data content in the database. Taking the telecommunications field as an example, user queries are generally divided into two categories: one is product and service data related to the user itself, such as "how much is my traffic left", "how much is my phone bill left", "my How many call minutes are left in my phone”, “How do I use my phone bill this month”; the other type is to inquire about th...

Embodiment 2

[0049] This embodiment provides an automatic question answering system, please refer to image 3 As shown, it includes a question interaction module 31, a question analysis module 32, and an answer processing module 33;

[0050] The question interaction module 31 is used to obtain the user question issued by the user;

[0051] The problem analysis module 32 is used for encoding the user's problem into a feature vector;

[0052] The answer processing module 33 is used to extract the answer to the user's question from the answer data of the path corresponding to the weight vector when one of the weight vectors in the feature vector and the path and weight correspondence table matches successfully The user; a weight vector corresponds to a path, and a path corresponds to a type of answer data; the answer data is data extracted from product and service data.

[0053] Specifically, the answer processing module 33 includes an answer mapping submodule 331 and an answer processing s...

example 1

[0076] Example 1, automatic question answering process:

[0077] Input: Question from Li Si (13012345002): "Check the phone bill."

[0078] Step 1, the question interaction module initializes the question, and the processing result is: "Check the phone bill."

[0079] Step 2: The processing result of the problem analysis module is: the feature vector of the problem is .

[0080]Step 3: The answer mapping module uses the feature vector of the question as to calculate the cosine similarity with each path vector. The preset similarity threshold is 0.95. The calculation results of each path are as follows:

[0081] User.Call balance: :

[0082]

[0083] User. Call bill details: :

[0084]

[0085] User.traffic balance:

[0086]

[0087] User.Traffic Details :

[0088]

[0089] Telecom building blocks. Flags :

[0090]

[0091] Telecom building blocks. Package content :

[0092]

[0093] Telecom building blocks. Handling process :

[0094]

[0095] Tel...

Images