CN111178489B - Conversation robot engine flow distribution method and device - Google Patents
Conversation robot engine flow distribution method and device Download PDFInfo
- Publication number
- CN111178489B CN111178489B CN201911398366.8A CN201911398366A CN111178489B CN 111178489 B CN111178489 B CN 111178489B CN 201911398366 A CN201911398366 A CN 201911398366A CN 111178489 B CN111178489 B CN 111178489B
- Authority
- CN
- China
- Prior art keywords
- flow
- training
- engine
- distribution
- result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/008—Artificial life, i.e. computing arrangements simulating life based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/70—Admission control; Resource allocation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L47/00—Traffic control in data switching networks
- H04L47/70—Admission control; Resource allocation
- H04L47/80—Actions related to the user profile or the type of traffic
Abstract
The application discloses a flow distribution method and device for an engine of a conversation robot, which are used for quickly and accurately distributing the engine flow to a more appropriate conversation robot engine, so that the conversation effect of the conversation robot is improved, and further the user experience is improved. The method in the embodiment of the application comprises the following steps: after an access conversation request input by a user is obtained, the access conversation request is firstly used as a flow to be distributed, then a flow parameter of the flow to be distributed is determined, and further reasonable distribution can be carried out on the flow to be distributed by utilizing a pre-constructed conversation robot engine flow distribution gateway according to the flow parameter to obtain a distribution result.
Description
Technical Field
The application relates to the technical field of computers, in particular to a conversation robot engine flow distribution method and device.
Background
With the rapid development of network technology and the continuous progress of robot technology, robots are increasingly popularized in daily life of people, and the robots can replace human beings to perform partial work so as to serve the human beings, thereby bringing convenience to the human beings. For example, the conversation robot can be used as an online customer service to provide convenient and fast conversation service for users.
At present, there are many manufacturers providing conversation robot engine services in the market, and due to the influence of complex factors such as corpus data, conversation objects, conversation scenes, and the like, different engine services can cause the conversation robot to generate different conversation effects. Therefore, how to quickly and accurately select a suitable engine service of the conversation robot to realize more reasonable engine flow distribution to the conversation robots of different engines, and improve the conversation effect of the conversation robot, has become a problem to be solved at present.
Disclosure of Invention
The embodiment of the application mainly aims to provide a method and a device for allocating engine flow of a conversation robot, which can quickly and accurately allocate the engine flow to a more appropriate conversation robot engine, improve the conversation effect of the conversation robot, and further improve user experience.
In a first aspect, an embodiment of the present application provides a method for allocating flow to an engine of a conversation robot, including:
acquiring flow to be distributed; the flow to be distributed is a dialogue access request input by a user;
determining a flow parameter of the flow to be distributed;
distributing the flow to be distributed by utilizing a pre-constructed flow distribution gateway of the conversation robot engine according to the flow parameters to obtain a distribution result;
constructing the flow distribution gateway of the conversation robot engine, comprising:
acquiring training flow;
determining a traffic parameter of the training traffic;
and training an initial dialogue robot engine flow distribution gateway according to the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to generate the dialogue robot engine flow distribution gateway.
In one possible implementation, the initial dialogue robot engine traffic distribution gateway includes an artificial neural network model; the artificial neural network model includes a fully connected layer.
In one possible implementation manner, training an initial dialogue robot engine traffic distribution gateway according to a traffic parameter of the training traffic and a dialogue robot distribution label corresponding to the training traffic to generate the dialogue robot engine traffic distribution gateway includes:
training an initial dialogue robot engine flow distribution gateway by using the flow parameter of the training flow and the dialogue robot distribution label corresponding to the training flow to obtain a training result;
according to the dialogue logs of the training flow, carrying out satisfaction scoring on the training result to obtain a scoring result;
and generating the flow distribution gateway of the conversation robot engine according to the grading result.
In a possible implementation, the method further includes:
obtaining verification flow;
determining a flow parameter of the verification flow;
inputting the flow parameters of the verification flow into the dialogue robot engine flow distribution gateway to obtain the dialogue robot distribution result of the verification flow;
and when the conversation robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow, the verification flow is used as the training flow again, and the conversation robot engine flow distribution gateway is updated.
In a second aspect, an embodiment of the present application further provides a conversation robot engine flow distribution device, including:
the first acquisition unit is used for acquiring the flow to be distributed; the flow to be distributed is a dialogue access request input by a user;
the first determining unit is used for determining the flow parameters of the flow to be distributed;
the distribution unit is used for distributing the flow to be distributed by utilizing a pre-constructed flow distribution gateway of the conversational robot engine according to the flow parameters to obtain a distribution result;
the second acquisition unit is used for acquiring the training flow;
a second determining unit, configured to determine a traffic parameter of the training traffic;
and the generating unit is used for training the initial dialogue robot engine flow distribution gateway according to the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow, and generating the dialogue robot engine flow distribution gateway.
In one possible implementation, the initial dialogue robot engine traffic distribution gateway includes an artificial neural network model; the artificial neural network model includes a fully connected layer.
In a possible implementation manner, the generating unit includes:
the training subunit is used for training an initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result;
the scoring subunit is used for scoring the satisfaction degree of the training result according to the dialogue log of the training flow to obtain a scoring result;
and the generating subunit is used for generating the conversation robot engine flow distribution gateway according to the grading result.
In a possible implementation manner, the apparatus further includes:
a third obtaining unit, configured to obtain a verification flow;
a third determining unit, configured to determine a flow parameter of the verification flow;
the input unit is used for inputting the flow parameters of the verification flow into the dialogue robot engine flow distribution gateway to obtain the dialogue robot distribution result of the verification flow;
and the updating unit is used for taking the verification flow as the training flow again and updating the dialogue robot engine flow distribution gateway when the dialogue robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow.
The embodiment of the application also provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on the terminal device, the terminal device is enabled to execute the above flow allocation method for the conversational robot engine.
The embodiment of the present application further provides a conversation robot engine flow distribution device, including: the flow distribution method comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the computer program, the flow distribution method of the dialogue robot engine is realized.
According to the method and the device for allocating the engine traffic of the conversation robot, after the access conversation request input by the user is obtained, the access conversation request is firstly used as the traffic to be allocated, then the traffic parameter of the traffic to be allocated is determined, and then the traffic to be allocated can be reasonably allocated by utilizing a pre-constructed conversation robot engine traffic allocation gateway according to the traffic parameter, so that an allocation result is obtained. It can be seen that, in the embodiment of the application, the pre-trained conversation robot engine flow distribution gateway is used for distributing the flow to be distributed input by the user, and determining that the appropriate conversation robot engine service is in conversation with the user, compared with manual conversation request distribution, the method for determining the type of the engine service of the conversation robot can realize automatic and rapid distribution of the flow to be distributed, and determine the more appropriate conversation robot engine service, and the distribution result eliminates the influence caused by the subjectivity of manual distribution, and has higher accuracy, so that the conversation effect of the conversation robot can be improved, and further the user experience is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic flowchart of a flow distribution method for an engine of a conversation robot according to an embodiment of the present disclosure;
fig. 2 is a schematic flowchart of a process for constructing a conversational robot engine traffic distribution gateway according to an embodiment of the present application;
fig. 3 is a schematic flowchart of a method for verifying a session robot engine traffic distribution gateway according to an embodiment of the present application;
fig. 4 is a diagram illustrating an example of a specific implementation of a method for allocating flow to an engine of a conversation robot according to an embodiment of the present disclosure;
fig. 5 is a second exemplary diagram of a specific implementation of a method for allocating flow to an engine of a conversation robot according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a conversation robot engine flow distribution device according to an embodiment of the present application.
Detailed Description
At present, there are many manufacturers providing conversation robot engine services in the market, and because the concurrent line capacity and the access amount guarantee/capping threshold of conversation robots produced by each manufacturer are different, different engine services can cause the conversation robots to generate different conversation effects, and in addition, the influence of characteristic factors such as language type and common conversation is added, for example, if a user is a conversation access request provided by a dialect, the conversation effect of the conversation robot with dialect capability will have a better conversation effect compared with other types of conversation robots. Based on this, in the prior art, for a dialog access request made by a user, it is usually necessary to manually select an engine service type of a corresponding dialog robot, so as to allocate the dialog access request to the dialog robot with a better dialog effect, so as to improve user experience. However, this manual determination method is highly subjective, difficult to quantify, and has low allocation efficiency. Therefore, how to quickly and accurately select a suitable engine service of the conversation robot to achieve more reasonable engine flow distribution to the conversation robots of different engines, and improve the distribution efficiency and the conversation effect of the conversation robot, has become a problem to be solved.
In order to solve the above defect, an embodiment of the present application provides a method for allocating a session robot engine traffic, where after an access session request input by a user is obtained, the access session request is first used as a traffic to be allocated, then a traffic parameter of the traffic to be allocated is determined, and then according to the traffic parameter, a pre-constructed session robot engine traffic allocation gateway is used to perform reasonable allocation on the traffic to be allocated, so as to obtain an allocation result. It can be seen that, in the embodiment of the application, the pre-trained conversation robot engine flow distribution gateway is used for distributing the flow to be distributed input by the user, and determining that the appropriate conversation robot engine service is in conversation with the user, compared with manual conversation request distribution, the method for determining the type of the engine service of the conversation robot can realize automatic and rapid distribution of the flow to be distributed, and determine the more appropriate conversation robot engine service, and the distribution result eliminates the influence caused by the subjectivity of manual distribution, and has higher accuracy, so that the conversation effect of the conversation robot can be improved, and further the user experience is improved.
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
First embodiment
Referring to fig. 1, a schematic flow chart of a method for allocating flow to an engine of a conversation robot according to this embodiment is provided, where the method includes the following steps:
s101: and acquiring the flow to be distributed, wherein the flow to be distributed is a dialogue access request input by a user.
In this embodiment, a flow of a session robot engine that needs to be allocated is defined as a flow to be allocated, where the flow to be allocated refers to a session access request input by a user, and the flow to be allocated may be allocated to a session robot service engine with a good session effect by the manner described in this embodiment, so as to improve user experience. It should be noted that, the embodiment does not limit the obtaining manner of the traffic to be allocated, for example, the traffic to be allocated may be obtained by receiving a voice message input by a user on the terminal device, or obtained by receiving a text message input by the user on the terminal device.
S102: and determining the flow parameters of the flow to be distributed.
In this embodiment, after the traffic to be distributed is obtained in step S101, the existing or future parameter extraction method may be used to process the acquired traffic to be distributed, so as to determine a parameter that can represent content information of the traffic to be distributed, where the parameter is defined as a traffic parameter. The traffic parameter may include information such as an access channel and a mobile phone number attribution, and for example, assuming that a terminal used by the user to input the traffic to be allocated is a mobile phone, the mobile phone number attribution used by the user to input the traffic to be allocated may be determined as a traffic parameter by using a mobile phone positioning method.
In an implementation manner of this embodiment, the traffic parameter may be extracted from the corresponding traffic to be distributed by using a pre-constructed dialogue robot engine traffic distribution gateway.
In this implementation manner, after the flow to be distributed is obtained in step S101, the flow cannot be directly distributed, but a flow parameter of the flow to be distributed needs to be determined first, and then the flow parameter is used to perform rapid and accurate distribution on the flow to be distributed by executing subsequent step S103, so as to determine the dialog robot engine with a better dialog effect.
S103: and distributing the flow to be distributed by utilizing a pre-constructed flow distribution gateway of the conversation robot engine according to the flow parameters to obtain a distribution result.
In this embodiment, after the flow parameters of the flow to be distributed are determined through step S102, the flow parameters may be input as input data into a pre-constructed flow distribution gateway of the dialog robot engine, so as to determine the dialog robot engine corresponding to the flow to be distributed through the gateway, thereby quickly and accurately distributing the engine flow to the corresponding more appropriate dialog robot engine, improving the dialog effect of the dialog robot, and further improving user experience.
Specifically, after determining the flow parameter of the flow to be distributed through step S102, the flow parameter may be input into the conversational robot engine flow distribution gateway to output a set of vectors representing the conversational robot engine classification, where a value of each dimension in the vectors may be a value in the interval [0,1], and the value of each dimension respectively represents a probability value that the flow to be distributed may be distributed to each preset conversational robot engine type. At this time, the dialog robot engine type corresponding to the maximum probability value can be used as the dialog robot engine for allocating the flow to be allocated.
For example, the following steps are carried out: it is assumed that there are 3 preset dialog robot engine types, namely, a dialog robot engine a, a dialog robot engine B, and a dialog robot engine C, and it is assumed that an output vector of the model is s ═ 0.9,0.1,0.05, and it is apparent that a value of the first dimension is 0.9 highest, and therefore, the dialog robot engine type corresponding to the dimension is the dialog robot engine for allocating traffic, that is, the dialog robot engine a can be selected as an online customer service to have a dialog with a user.
Alternatively, a probability threshold value for classifying the conversational robot engines may be preset, and one robot engine type with an output probability greater than the threshold value may be used as the conversational robot engine to be allocated for the traffic to be allocated.
For example, the following steps are carried out: based on the above example, it is assumed that there are three preset conversation robot engine types, namely "conversation robot engine 1, conversation robot engine 2, and conversation robot engine 3", and the probability threshold of the conversation robot engine classification is set to 0.6 in advance, and after the conversation robot engine traffic distribution gateway distributes the data, the output vector is "s ═ 0.2,0.8, 0.1", and it can be seen that the value of the second dimension 0.8 exceeds the probability threshold of the preset conversation robot engine classification, and therefore, the conversation robot engine type corresponding to the dimension can be used as the conversation robot engine to be distributed with traffic, that is, the conversation robot engine 2 can be selected as an online customer service to have a conversation with the user.
It should be noted that, in the embodiment of the present application, each preset different type of session robot engine needs to access the same database, that is, it is required to ensure that the indication data of each preset different type of session robot engine are consistent.
Further, after the conversation robot engine of the flow distribution to be distributed is determined, namely after the dynamic distribution of the flow to be distributed is realized, the conversation effect of each type of conversation robot engine can be evaluated according to the flow distribution quantity of each conversation robot engine, so that the conversation robot engine service type with a better conversation effect is selected according to the evaluation result, and the investment return rate of an enterprise is improved.
It should be noted that, to implement this step S103, a conversation robot engine traffic distribution gateway needs to be constructed in advance, and the specific construction process can be referred to in the related description of the second embodiment.
In summary, according to the method for allocating the engine traffic of the conversation robot provided in this embodiment, after the access conversation request input by the user is obtained, the access conversation request is first used as the traffic to be allocated, then the traffic parameter of the traffic to be allocated is determined, and then the traffic to be allocated can be reasonably allocated by using the pre-constructed conversation robot engine traffic allocation gateway according to the traffic parameter, so as to obtain the allocation result. It can be seen that, in the embodiment of the application, the pre-trained conversation robot engine flow distribution gateway is used for distributing the flow to be distributed input by the user, and determining that the appropriate conversation robot engine service is in conversation with the user, compared with manual conversation request distribution, the method for determining the type of the engine service of the conversation robot can realize automatic and rapid distribution of the flow to be distributed, and determine the more appropriate conversation robot engine service, and the distribution result eliminates the influence caused by the subjectivity of manual distribution, and has higher accuracy, so that the conversation effect of the conversation robot can be improved, and further the user experience is improved.
Second embodiment
The present embodiment will describe a specific construction process of the dialogue robot engine traffic distribution gateway mentioned in the first embodiment. By utilizing the pre-constructed flow distribution gateway of the conversation robot engine, the conversation robot engine corresponding to the flow to be distributed can be accurately and quickly determined.
Referring to fig. 2, it shows a schematic flow chart of building a conversational robot engine traffic distribution gateway according to this embodiment, where the flow chart includes the following steps:
s201: and acquiring training flow.
In this embodiment, in order to construct the session robot engine traffic distribution gateway, a large amount of preparation work needs to be performed in advance, and first, training traffic needs to be obtained, for example, 100 session access requests input by users may be collected in advance as training traffic, each collected training traffic is respectively used as sample traffic data, and the session robot engine types corresponding to the sample traffic data are marked in advance by a human to train the session robot engine traffic distribution gateway.
S202: a flow parameter of the training flow is determined.
In this embodiment, after the training traffic is obtained in step S201, the training traffic cannot be directly used for training to generate the session robot engine traffic distribution gateway, but traffic parameters of the training traffic, such as a home location and an access channel of a mobile phone number corresponding to a session access request input by a user, are determined first. In the parameter determining process, the existing parameter extraction method can be used for extraction, the dialogue robot engine flow distribution gateway constructed in the embodiment can also be used for extraction, and then the extracted flow parameters of the training flow can be used for training to obtain the dialogue robot engine flow distribution gateway.
S203: and training the initial dialogue robot engine flow distribution gateway according to the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to generate the dialogue robot engine flow distribution gateway.
In this embodiment, after the flow parameter of the training flow is determined in step S202, the dialog robot engine flow allocation gateway may be further trained according to the flow parameter of the training flow and the labeling result of the dialog robot engine type corresponding to the training flow, so as to generate the dialog robot engine flow allocation gateway. In an optional implementation manner, the initial dialogue robot engine traffic distribution gateway includes an artificial neural network model, and the artificial neural network model includes a full connection layer.
Specifically, in an optional implementation manner of the embodiment of the present application, the implementation process of this step S203 may specifically include the following steps a to C:
step A: and training the initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result.
In this implementation, the flow parameter of the training flow is defined as s, and the allocation result output by the initial dialogue robot engine flow allocation gateway is defined as a. And all pairs (s, a) acquired in a fixed period (e.g. one day or one week) form a "track", which is defined herein as τ, i.e. τ is (s, a), and τ may be used as the training result of the initial conversational robot engine traffic distribution gateway during this period.
And B: and according to the dialogue logs of the training flow, carrying out satisfaction scoring on the training result to obtain a scoring result.
In this implementation, after the initial dialogue robot engine traffic distribution gateway is trained by using the traffic parameter of the training traffic and the dialogue robot assignment tag corresponding to the training traffic to obtain the training result τ in step a, the effect index data after each training dialogue can be calculated according to the dialogue log of the training traffic (i.e., the training dialogue access request input by the historical user), and further, the training result of the initial dialogue robot engine traffic distribution gateway can be scored according to the difference between the historical effect index data and the manual annotation result corresponding to the training traffic to obtain a scoring result, which is represented by R (τ).
And C: and generating a flow distribution gateway of the conversation robot engine according to the grading result.
In this implementation manner, according to the result data obtained in the above steps, a policy gradient (policy gradient) algorithm may be used to train an artificial neural network model in the initial session robot engine traffic distribution gateway, so as to update the model parameters and improve the scoring result of the entire gateway until a preset condition is met, for example, if the variation range of the scoring result is small or the scoring result is substantially higher than a preset scoring threshold, the updating of the model parameters is stopped, the training of the session robot engine traffic distribution gateway is completed, and a trained session robot engine traffic distribution gateway is generated.
Specifically, the calculation formula of the satisfaction expectation in the above adopted strategy gradient algorithm is as follows:
wherein the content of the first and second substances,
representing a satisfaction expectation; n represents the total number of training results; r (tau) represents the satisfaction score of the training result tau; pθAnd (tau) represents the probability of the occurrence of the training result tau when the parameter of the artificial neural network model in the initial dialogue robot engine flow distribution gateway is theta.
In addition, in order to improve the satisfaction expectation value, gradient calculation needs to be performed on the satisfaction expectation value, and a specific calculation formula is as follows:
in practical applications, the above formula (2) can be converted into the following formula (3):
wherein, TnRepresents the length of the training result τ;
a flow parameter indicating a training flow at the t-th time in the training result τ;
representing the distribution result output by the engine flow distribution gateway of the initial dialogue robot at the t-th moment in the training result tau;
indicating that the current initial dialog robot engine flow distribution gateway is at the input
Under the condition of (1), output
The conditional probability of (2).
Further, the model parameters may be updated according to the following equation (4):
wherein, thetanRepresenting model parameters at time n; thetan-1Representing model parameters at the n-1 th moment; η represents an updated step length, and a specific value may be determined according to an actual situation and an empirical value, which is not limited in the embodiment of the present application, for example, η may be taken as 10-4Or 10-5And the like.
Through the embodiment, the dialogue robot engine traffic distribution gateway can be generated by training the training traffic, and further, the generated dialogue robot engine traffic distribution gateway can be verified by using the verification traffic.
Referring to fig. 3, which shows a schematic flowchart of a method for verifying a session robot engine traffic distribution gateway according to an embodiment of the present application, as shown in fig. 3, the method includes:
s301: and obtaining the verification flow.
In practical applications, in order to implement the verification of the session robot engine traffic distribution gateway, first, a verification traffic needs to be obtained, where the verification traffic refers to an engine traffic (i.e., a session access request input by a user) that can be used for performing the verification of the session robot engine traffic distribution gateway, and after the verification traffic is obtained, S302 may be continuously performed.
S302: a flow parameter is determined that validates the flow.
In practical application, after the verification traffic is obtained in step S301, the verification traffic cannot be directly used for verifying and identifying the conversation robot engine traffic distribution gateway, but the traffic parameters of the verification traffic need to be extracted, for example, the mobile phone number attribution and the access channel corresponding to the conversation access request input by the user. In the parameter determination process, the existing parameter extraction method can be used for extraction, the dialogue robot engine flow distribution gateway constructed in the embodiment can also be used for extraction, and further the extracted flow parameter for verifying the flow can be used for verifying the obtained dialogue robot engine flow distribution gateway.
S303: and inputting the flow parameters of the verification flow into the flow distribution gateway of the dialogue robot engine to obtain the dialogue robot distribution result of the verification flow.
In a specific implementation process, after the flow parameter of the verification flow is determined in step S302, the flow parameter of the verification flow may be further input to the conversational robot engine flow distribution gateway to obtain a conversational robot distribution result of the verification flow, and then step S304 may be continuously performed.
S304: and when the conversation robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow, the verification flow is used as the training flow again, and the conversation robot engine flow distribution gateway is updated.
In practical application, after the conversation robot distribution result of the verification traffic is obtained in step S303, when the conversation robot distribution result of the verification traffic is inconsistent with the manual labeling result corresponding to the verification traffic, the verification traffic may be used as the training traffic again to update the conversation robot engine traffic distribution gateway.
Through the embodiment, the dialog robot engine flow distribution gateway can be effectively verified by utilizing the verification flow, and when the dialog robot distribution result of the verification flow is inconsistent with the manual marking result corresponding to the verification flow, the dialog robot engine flow distribution gateway can be timely adjusted and updated, so that the distribution precision and accuracy of the dialog robot engine flow distribution gateway can be improved.
In summary, the dialogue robot engine traffic distribution gateway trained by the embodiment can quickly and accurately distribute the engine traffic to the corresponding more appropriate dialogue robot engine by using the traffic parameter representing the content information of the traffic to be distributed, and in the distribution process, the calculated amount of the parameter is greatly reduced, and the efficiency and the accuracy of the distribution result are effectively improved.
Third embodiment
For the convenience of understanding, the present embodiment is combined with a specific implementation example diagram of the conversation robot engine flow allocation method shown in fig. 4. The implementation process of the conversation robot engine flow distribution method provided by the embodiment of the application is introduced.
As shown in fig. 4, in this example, the dialogue robot engine traffic distribution gateway includes: the system comprises an access and flow distribution module, a real-time decision module and an evaluation module. The implementation process of the embodiment of the application is as follows: firstly, after an access and flow distribution module obtains an access session request (i.e., the robot application 1, the robot application 2, or the robot application 3 in fig. 4) input by a user, it first takes the access and flow distribution request as a flow to be distributed, then determines flow parameters of the flow to be distributed, such as information of an access channel, a mobile phone number attribution, and the like, and then sends the information to a real-time decision module, and then, according to the flow parameters and historical effect data provided by an evaluation module, reasonably distributes the flow to be distributed by using a neural network model pre-constructed in the real-time decision module, and determines a more appropriate session robot engine service (i.e., the robot engine 1, the robot engine 2, or the robot engine 3 in fig. 4), and the specific implementation process refers to steps S101 to S103.
For the convenience of understanding, the present embodiment is further combined with a specific implementation example diagram of the conversation robot engine traffic distribution method shown in fig. 5. The implementation process of the conversation robot engine flow distribution method provided by the embodiment of the application is introduced.
As shown in fig. 5, in this example, the dialogue robot engine traffic distribution gateway includes: a distribution decision module and an effect evaluation module. The implementation process of the embodiment of the application is as follows: firstly, after an access conversation request input by a user is accessed through a telephone, the access conversation request is used as a flow to be distributed so as to determine flow parameters of the flow to be distributed, such as information of an access channel, a mobile phone number attribution and the like, and then the information is sent to a distribution decision module, so that the flow to be distributed can be reasonably distributed by utilizing a neural network model pre-constructed in the distribution decision module according to the flow parameters and historical effect data provided by an effect evaluation module, and more appropriate conversation robot engine services (namely, a telephone robot 1, a telephone robot 2 or a telephone robot 3 in fig. 4) are determined, and the specific implementation process is shown in steps S101 to S103.
Fourth embodiment
In this embodiment, a flow distribution device for an engine of a conversation robot will be described, and please refer to the above method embodiment for related contents.
Referring to fig. 6, a schematic composition diagram of a conversation robot engine flow distribution device provided in this embodiment is shown, where the device includes:
a first obtaining unit 601, configured to obtain a flow to be allocated; the flow to be distributed is a dialogue access request input by a user;
a first determining unit 602, configured to determine a flow parameter of the flow to be allocated;
the allocating unit 603 is configured to allocate the traffic to be allocated according to the traffic parameter by using a pre-established session robot engine traffic allocation gateway, so as to obtain an allocation result;
a second obtaining unit 604, configured to obtain a training flow;
a second determining unit 605, configured to determine a traffic parameter of the training traffic;
a generating unit 606, configured to train an initial dialogue robot engine traffic distribution gateway according to the traffic parameter of the training traffic and the dialogue robot distribution label corresponding to the training traffic, and generate the dialogue robot engine traffic distribution gateway.
In one implementation manner of this embodiment, the initial dialogue robot engine traffic distribution gateway includes an artificial neural network model; the artificial neural network model includes a fully connected layer.
In an implementation manner of this embodiment, the generating unit 606 includes:
the training subunit is used for training an initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result;
the scoring subunit is used for scoring the satisfaction degree of the training result according to the dialogue log of the training flow to obtain a scoring result;
and the generating subunit is used for generating the conversation robot engine flow distribution gateway according to the grading result.
In an implementation manner of this embodiment, the apparatus further includes:
a third obtaining unit, configured to obtain a verification flow;
a third determining unit, configured to determine a flow parameter of the verification flow;
the input unit is used for inputting the flow parameters of the verification flow into the dialogue robot engine flow distribution gateway to obtain the dialogue robot distribution result of the verification flow;
and the updating unit is used for taking the verification flow as the training flow again and updating the dialogue robot engine flow distribution gateway when the dialogue robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow.
In summary, after the access session request input by the user is obtained, the session robot engine traffic distribution device provided in this embodiment first uses the access session request as the traffic to be distributed, then determines the traffic parameter of the traffic to be distributed, and further may use a pre-constructed session robot engine traffic distribution gateway to perform reasonable distribution on the traffic to be distributed according to the traffic parameter, so as to obtain a distribution result. It can be seen that, in the embodiment of the application, the pre-trained conversation robot engine flow distribution gateway is used for distributing the flow to be distributed input by the user, and determining that the appropriate conversation robot engine service is in conversation with the user, compared with manual conversation request distribution, the method for determining the type of the engine service of the conversation robot can realize automatic and rapid distribution of the flow to be distributed, and determine the more appropriate conversation robot engine service, and the distribution result eliminates the influence caused by the subjectivity of manual distribution, and has higher accuracy, so that the conversation effect of the conversation robot can be improved, and further the user experience is improved.
In addition, a computer-readable storage medium is provided, where instructions are stored, and when the instructions are run on a terminal device, the terminal device is caused to execute the above method for allocating conversational robot engine traffic.
An embodiment of the present application further provides a data processing apparatus, including: the flow distribution method comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the computer program, the flow distribution method of the dialogue robot engine is realized.
The embodiment of the application also provides a computer program product, and when the computer program product runs on the terminal equipment, the terminal equipment executes the conversation robot engine flow distribution method.
When introducing elements of various embodiments of the present application, the articles "a," "an," "the," and "said" are intended to mean that there are one or more of the elements. The terms "comprising," "including," and "having" are intended to be inclusive and mean that there may be additional elements other than the listed elements.
It should be noted that, as one of ordinary skill in the art would understand, all or part of the processes of the above method embodiments may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when executed, the computer program may include the processes of the above method embodiments. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus embodiment, since it is substantially similar to the method embodiment, it is relatively simple to describe, and reference may be made to some descriptions of the method embodiment for relevant points. The above-described apparatus embodiments are merely illustrative, and the units and modules described as separate components may or may not be physically separate. In addition, some or all of the units and modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The foregoing is directed to embodiments of the present application and it is noted that numerous modifications and adaptations may be made by those skilled in the art without departing from the principles of the present application and are intended to be within the scope of the present application.
Claims (10)
1. A conversation robot engine flow distribution method is characterized by comprising the following steps:
acquiring flow to be distributed; the flow to be distributed is a dialogue access request input by a user;
determining a flow parameter of the flow to be distributed;
distributing the flow to be distributed by utilizing a pre-constructed flow distribution gateway of the conversation robot engine according to the flow parameters to obtain a distribution result; after determining the flow parameters of the flow to be distributed, inputting the flow parameters into a flow distribution gateway of an engine of the conversation robot so as to output a group of vectors representing the classification of the engine of the conversation robot, wherein the value of each dimension in the vectors is a numerical value in an interval [0,1], and the value of each dimension respectively represents the probability value of the flow to be distributed to each preset conversation robot engine type; at the moment, the type of the conversation robot engine corresponding to the maximum probability value is used as the conversation robot engine for distributing the flow to be distributed;
constructing the flow distribution gateway of the conversation robot engine, comprising:
acquiring training flow;
determining a traffic parameter of the training traffic;
training an initial dialogue robot engine flow distribution gateway according to the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to generate the dialogue robot engine flow distribution gateway; the method specifically comprises the following steps A-C:
step A: training an initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result;
defining the flow parameter of the training flow as s, and defining the distribution result output by the engine flow distribution gateway of the initial dialogue robot as a; all pairs (s, a) acquired in a fixed period form a "track", which is defined as τ, that is, τ is (s, a), and τ is used as a training result of the traffic distribution gateway of the initial conversational robot engine in the period;
and B: according to the dialogue logs of the training flow, carrying out satisfaction scoring on the training result to obtain a scoring result;
a, training an initial dialogue robot engine flow distribution gateway by using a flow parameter of training flow and a dialogue robot distribution label corresponding to the training flow to obtain a training result tau, calculating effect index data after each training dialogue according to a dialogue log of the training flow, and grading the training result of the initial dialogue robot engine flow distribution gateway according to the historical effect index data and the difference between manual labeling results corresponding to the training flow to obtain a grading result, wherein the grading result is expressed by R (tau);
and C: generating a flow distribution gateway of the conversation robot engine according to the grading result;
training an artificial neural network model in the initial dialogue robot engine flow distribution gateway by using a strategy gradient algorithm to update model parameters, improving the grading result of the whole gateway, stopping updating of the model parameters until a preset condition is met, completing training of the dialogue robot engine flow distribution gateway, and generating a trained dialogue robot engine flow distribution gateway;
the calculation formula of the satisfaction expectation in the strategy gradient algorithm is as follows:
wherein the content of the first and second substances,
representing a satisfaction expectation; n represents the total number of training results; r (tau) represents the satisfaction score of the training result tau; pθ(tau) represents the probability of the occurrence of a training result tau when the parameter of an artificial neural network model in the flow distribution gateway of the initial dialogue robot engine is theta;
and performing gradient calculation on the satisfaction expectation, wherein the specific calculation formula is as follows:
converting the above formula (2) into the following formula (3):
wherein, TnRepresents the length of the training result τ;
a flow parameter indicating a training flow at the t-th time in the training result τ;
representing the distribution result output by the engine flow distribution gateway of the initial dialogue robot at the t-th moment in the training result tau;
indicating that the current initial dialog robot engine flow distribution gateway is at the input
Under the condition of (1), output
The conditional probability of (a);
updating the model parameters according to the following formula (4):
wherein, thetanRepresenting model parameters at time n; thetan-1Representing model parameters at the n-1 th moment; η represents the step size of the update.
2. The method of claim 1, wherein the initial dialogue robot engine traffic distribution gateway comprises an artificial neural network model; the artificial neural network model includes a fully connected layer.
3. The method according to any one of claims 1 to 2, wherein training an initial dialogue robot engine traffic distribution gateway according to the traffic parameter of the training traffic and a dialogue robot distribution label corresponding to the training traffic to generate the dialogue robot engine traffic distribution gateway comprises:
training an initial dialogue robot engine flow distribution gateway by using the flow parameter of the training flow and the dialogue robot distribution label corresponding to the training flow to obtain a training result;
according to the dialogue logs of the training flow, carrying out satisfaction scoring on the training result to obtain a scoring result;
and generating the flow distribution gateway of the conversation robot engine according to the grading result.
4. The method of claim 1, further comprising:
obtaining verification flow;
determining a flow parameter of the verification flow;
inputting the flow parameters of the verification flow into the dialogue robot engine flow distribution gateway to obtain the dialogue robot distribution result of the verification flow;
and when the conversation robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow, the verification flow is used as the training flow again, and the conversation robot engine flow distribution gateway is updated.
5. A conversational robot engine flow distribution device, comprising:
the first acquisition unit is used for acquiring the flow to be distributed; the flow to be distributed is a dialogue access request input by a user;
the first determining unit is used for determining the flow parameters of the flow to be distributed;
the distribution unit is used for distributing the flow to be distributed by utilizing a pre-constructed flow distribution gateway of the conversational robot engine according to the flow parameters to obtain a distribution result; after determining the flow parameters of the flow to be distributed, inputting the flow parameters into a flow distribution gateway of an engine of the conversation robot so as to output a group of vectors representing the classification of the engine of the conversation robot, wherein the value of each dimension in the vectors is a numerical value in an interval [0,1], and the value of each dimension respectively represents the probability value of the flow to be distributed to each preset conversation robot engine type; at the moment, the type of the conversation robot engine corresponding to the maximum probability value is used as the conversation robot engine for distributing the flow to be distributed;
the second acquisition unit is used for acquiring the training flow;
a second determining unit, configured to determine a traffic parameter of the training traffic;
the generating unit is used for training an initial dialogue robot engine flow distribution gateway according to the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to generate the dialogue robot engine flow distribution gateway; the method specifically comprises the following steps A-C:
step A: training an initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result;
defining the flow parameter of the training flow as s, and defining the distribution result output by the engine flow distribution gateway of the initial dialogue robot as a; all pairs (s, a) acquired in a fixed period form a "track", which is defined as τ, that is, τ is (s, a), and τ is used as a training result of the traffic distribution gateway of the initial conversational robot engine in the period;
and B: according to the dialogue logs of the training flow, carrying out satisfaction scoring on the training result to obtain a scoring result;
a, training an initial dialogue robot engine flow distribution gateway by using a flow parameter of training flow and a dialogue robot distribution label corresponding to the training flow to obtain a training result tau, calculating effect index data after each training dialogue according to a dialogue log of the training flow, and grading the training result of the initial dialogue robot engine flow distribution gateway according to the historical effect index data and the difference between manual labeling results corresponding to the training flow to obtain a grading result, wherein the grading result is expressed by R (tau);
and C: generating a flow distribution gateway of the conversation robot engine according to the grading result;
training an artificial neural network model in the initial dialogue robot engine flow distribution gateway by using a strategy gradient algorithm to update model parameters, improving the grading result of the whole gateway, stopping updating of the model parameters until a preset condition is met, completing training of the dialogue robot engine flow distribution gateway, and generating a trained dialogue robot engine flow distribution gateway;
the calculation formula of the satisfaction expectation in the strategy gradient algorithm is as follows:
wherein the content of the first and second substances,
representing a satisfaction expectation; n represents the total number of training results; r (tau) represents the satisfaction score of the training result tau; pθ(tau) represents the probability of the occurrence of a training result tau when the parameter of an artificial neural network model in the flow distribution gateway of the initial dialogue robot engine is theta;
and performing gradient calculation on the satisfaction expectation, wherein the specific calculation formula is as follows:
converting the above formula (2) into the following formula (3):
wherein, TnRepresents the length of the training result τ;
a flow parameter indicating a training flow at the t-th time in the training result τ;
representing the distribution result output by the engine flow distribution gateway of the initial dialogue robot at the t-th moment in the training result tau;
indicating that the current initial dialog robot engine flow distribution gateway is at the input
Under the condition of (1), output
The conditional probability of (a);
updating the model parameters according to the following formula (4):
wherein, thetanRepresenting model parameters at time n; thetan-1Representing model parameters at the n-1 th moment; η represents the step size of the update.
6. The apparatus of claim 5, wherein the initial dialogue robot engine traffic distribution gateway comprises an artificial neural network model; the artificial neural network model includes a fully connected layer.
7. The apparatus according to any one of claims 5 to 6, wherein the generating unit comprises:
the training subunit is used for training an initial dialogue robot engine flow distribution gateway by using the flow parameters of the training flow and the dialogue robot distribution labels corresponding to the training flow to obtain a training result;
the scoring subunit is used for scoring the satisfaction degree of the training result according to the dialogue log of the training flow to obtain a scoring result;
and the generating subunit is used for generating the conversation robot engine flow distribution gateway according to the grading result.
8. The apparatus of any of claims 5 to 6, further comprising:
a third obtaining unit, configured to obtain a verification flow;
a third determining unit, configured to determine a flow parameter of the verification flow;
the input unit is used for inputting the flow parameters of the verification flow into the dialogue robot engine flow distribution gateway to obtain the dialogue robot distribution result of the verification flow;
and the updating unit is used for taking the verification flow as the training flow again and updating the dialogue robot engine flow distribution gateway when the dialogue robot distribution result of the verification flow is inconsistent with the distribution marking result corresponding to the verification flow.
9. A computer-readable storage medium having stored therein instructions that, when run on a terminal device, cause the terminal device to perform the conversational robot engine traffic distribution method of any of claims 1-4.
10. A conversational robot engine flow distribution device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, when executing the computer program, implementing the conversational robot engine traffic distribution method of any of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911398366.8A CN111178489B (en) | 2019-12-30 | 2019-12-30 | Conversation robot engine flow distribution method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911398366.8A CN111178489B (en) | 2019-12-30 | 2019-12-30 | Conversation robot engine flow distribution method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111178489A CN111178489A (en) | 2020-05-19 |
| CN111178489B true CN111178489B (en) | 2021-02-19 |
Family
ID=70652275
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911398366.8A Active CN111178489B (en) | 2019-12-30 | 2019-12-30 | Conversation robot engine flow distribution method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111178489B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101076060A (en) * | 2007-03-30 | 2007-11-21 | 腾讯科技（深圳）有限公司 | Chatting robot system and automatic chatting method |
| CN106897263A (en) * | 2016-12-29 | 2017-06-27 | 北京光年无限科技有限公司 | Robot dialogue exchange method and device based on deep learning |
| CN108319599A (en) * | 2017-01-17 | 2018-07-24 | 华为技术有限公司 | A kind of interactive method and apparatus |
| CN108491394A (en) * | 2018-06-27 | 2018-09-04 | 杭州贝店科技有限公司 | A kind of semantic analysis, device, computer equipment and storage medium |
| CN108733722A (en) * | 2017-04-24 | 2018-11-02 | 北京京东尚科信息技术有限公司 | A kind of dialogue robot automatic generation method and device |
| CN110175227A (en) * | 2019-05-10 | 2019-08-27 | 神思电子技术股份有限公司 | A kind of dialogue auxiliary system based on form a team study and level reasoning |
| CN110362667A (en) * | 2019-07-12 | 2019-10-22 | 深圳前海微众银行股份有限公司 | Intelligent customer service method, apparatus, equipment and readable storage medium storing program for executing |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101821358B1 (en) * | 2013-01-22 | 2018-01-25 | 네이버 주식회사 | Method and system for providing multi-user messenger service |
| CN104123939A (en) * | 2014-06-06 | 2014-10-29 | 国家电网公司 | Substation inspection robot based voice interaction control method |
| CN106055662A (en) * | 2016-06-02 | 2016-10-26 | 竹间智能科技（上海）有限公司 | Emotion-based intelligent conversation method and system |
| CN109492854A (en) * | 2018-09-17 | 2019-03-19 | 平安科技（深圳）有限公司 | Method, apparatus, computer equipment and the storage medium of intelligent robot interview |
-
2019
- 2019-12-30 CN CN201911398366.8A patent/CN111178489B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101076060A (en) * | 2007-03-30 | 2007-11-21 | 腾讯科技（深圳）有限公司 | Chatting robot system and automatic chatting method |
| CN106897263A (en) * | 2016-12-29 | 2017-06-27 | 北京光年无限科技有限公司 | Robot dialogue exchange method and device based on deep learning |
| CN108319599A (en) * | 2017-01-17 | 2018-07-24 | 华为技术有限公司 | A kind of interactive method and apparatus |
| CN108733722A (en) * | 2017-04-24 | 2018-11-02 | 北京京东尚科信息技术有限公司 | A kind of dialogue robot automatic generation method and device |
| CN108491394A (en) * | 2018-06-27 | 2018-09-04 | 杭州贝店科技有限公司 | A kind of semantic analysis, device, computer equipment and storage medium |
| CN110175227A (en) * | 2019-05-10 | 2019-08-27 | 神思电子技术股份有限公司 | A kind of dialogue auxiliary system based on form a team study and level reasoning |
| CN110362667A (en) * | 2019-07-12 | 2019-10-22 | 深圳前海微众银行股份有限公司 | Intelligent customer service method, apparatus, equipment and readable storage medium storing program for executing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111178489A (en) | 2020-05-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106951468B (en) | Talk with generation method and device | |
| CN107886949A (en) | A kind of content recommendation method and device | |
| CN107330706A (en) | A kind of electricity battalion's customer service system and commercial operation pattern based on artificial intelligence | |
| CN110457709A (en) | Outgoing call dialog process method, apparatus and server | |
| CN110334201A (en) | A kind of intension recognizing method, apparatus and system | |
| CN109670023A (en) | Man-machine automatic top method for testing, device, equipment and storage medium | |
| CN108073600A (en) | A kind of intelligent answer exchange method, device and electronic equipment | |
| US20210150398A1 (en) | Conversational interchange optimization | |
| CN106897206A (en) | A kind of service test method and device | |
| CN106303113B (en) | A kind of shunt method and equipment | |
| CN106302579A (en) | The distribution method of a kind of task and server | |
| CN108198017B (en) | Personalized customer service processing method and device | |
| CN111858854B (en) | Question-answer matching method and relevant device based on historical dialogue information | |
| US10264128B1 (en) | Dynamic machine-learning-based contact processing | |
| CN110445939B (en) | Capacity resource prediction method and device | |
| CN106454437A (en) | Streaming media service rate prediction method and device | |
| CN111949795A (en) | Work order automatic classification method and device | |
| CN110019838A (en) | Intelligent Answer System and intelligent terminal | |
| CN109040489B (en) | Method and device for distributing telecommunication customer service | |
| CN107977676A (en) | Text similarity computing method and device | |
| CN107967304A (en) | Session interaction processing method, device and electronic equipment | |
| CN106897282A (en) | The sorting technique and equipment of a kind of customer group | |
| CN113254840B (en) | Artificial intelligence application service pushing method, pushing platform and terminal equipment | |
| CN111178489B (en) | Conversation robot engine flow distribution method and device | |
| CN115953080B (en) | Engineer service level determination method, apparatus and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |