CN110716767B - Model component calling and generating method, device and storage medium - Google Patents

Model component calling and generating method, device and storage medium Download PDF

Info

Publication number
CN110716767B
CN110716767B CN201810769756.0A CN201810769756A CN110716767B CN 110716767 B CN110716767 B CN 110716767B CN 201810769756 A CN201810769756 A CN 201810769756A CN 110716767 B CN110716767 B CN 110716767B
Authority
CN
China
Prior art keywords
model component
material set
specified
customized
customized model
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
Application number
CN201810769756.0A
Other languages
Chinese (zh)
Other versions
CN110716767A (en
Inventor
周躜
王凯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alibaba Group Holding Ltd
Original Assignee
Alibaba Group Holding Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Alibaba Group Holding Ltd filed Critical Alibaba Group Holding Ltd
Priority to CN201810769756.0A priority Critical patent/CN110716767B/en
Publication of CN110716767A publication Critical patent/CN110716767A/en
Application granted granted Critical
Publication of CN110716767B publication Critical patent/CN110716767B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44Arrangements for executing specific programs
    • G06F9/448Execution paradigms, e.g. implementations of programming paradigms
    • G06F9/4482Procedural

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the specification discloses a model component calling and generating method, a device and a storage medium. The scheme can realize automatic generation of the model component. Therefore, the model component generation process can be realized without further manual intervention, and the manual labor is reduced.

Description

Model component calling and generating method, device and storage medium
Technical Field
The present disclosure relates to the field of computer data processing, and in particular, to a method, an apparatus, and a storage medium for generating a model component.
Background
With the development of society, computer software is widely applied to a plurality of fields of society, and brings great convenience to the production and life of people. For example, the electrical device is controlled by audio recognition.
In the prior art, a certain service function can be realized by compiling a computer model component, the computer model component usually is a generating process of a model component which is manually selected by a developer according to use requirements, an algorithm is selected, parameters are set and the like, and the model component is adjusted through experience.
Disclosure of Invention
The embodiment of the specification provides a model component generating method, a device and a storage medium for generating a model component conveniently.
The embodiment of the specification provides a calling method of a model component, which comprises the following steps: providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; and calling the customized model component to realize the specified business function.
The embodiment of the specification provides a calling device of a model component, which comprises: a generation module that provides a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; and the calling module is used for calling the customized model component to realize the specified service function.
The present description provides a computer storage medium storing computer program instructions that, when executed, implement: providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; and calling the customized model component to realize the specified business function.
The embodiment of the specification provides a calling method of a model component, which comprises the following steps: receiving a material set designated by a user; wherein the material set comprises at least one material; transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; receiving a customized model component fed back by the server; and calling the customized model component to realize the specified service function.
The embodiment of the specification provides a calling device of a model component, which comprises: the first receiving module is used for receiving a material set appointed by a user; wherein the material set comprises at least one material; the sending module is used for sending the material set to a server, so that a machine learning model component of the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the second receiving module is used for receiving the customized model component fed back by the server; and the calling module is used for calling the customized model component to realize the specified service function.
The present description provides a computer storage medium storing computer program instructions that, when executed, implement: receiving a material set designated by a user; wherein the material set comprises at least one material; transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; receiving a customized model component fed back by the server; and calling the customized model component to realize the specified service function.
The embodiment of the specification provides a calling method of a model component, which comprises the following steps: transmitting the material set to a server for the server to take the material set as input to a machine learning model component and generate a customized model component based on a specified model component; and sending a call request to call the customized model component to realize the specified service function.
The embodiment of the specification provides a calling device of a model component, which comprises: the system comprises a material sending module, a machine learning model component, a customization model component and a material processing module, wherein the material sending module is used for sending a material set to a server, and the server is used for taking the material set as the input of the machine learning model component and generating the customization model component based on the specified model component; and the request sending module is used for sending a calling request to call the customized model component to realize the specified service function.
The embodiment of the specification provides a model component generating method, which comprises the following steps: determining a material set for generating a customized model component; wherein the material set comprises a plurality of materials; according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining a training process of the designated model component; and taking the material set as input, and obtaining the customized model component based on a training process of the specified model component according to the first constraint condition constraint.
The embodiment of the present specification provides a model component generating apparatus, including: a determining module for determining a material set for generating a customized model component; wherein the material set comprises a plurality of materials; the formulating module is used for formulating a first constraint condition of the appointed model component according to the material set, wherein the first constraint condition is used for constraining the training process of the appointed model component; and the generation module is used for taking the material set as input and generating the customized model component based on the training process of the specified model component according to the first constraint condition constraint.
The present description provides a computer storage medium storing computer program instructions that, when executed, implement: determining a set of materials for generating a customized model component, wherein the set of materials includes a plurality of materials; according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining a training process of the designated model component; and taking the material set as input, and obtaining the customized model component based on a training process of the specified model component according to the first constraint condition constraint.
From the above, the technical solution provided in the present specification makes constraint conditions by analyzing the target material set to constrain the generation process of the model component. Therefore, after the target material set is specified, the model component can be automatically generated without further manual intervention. The manual labor is reduced. In addition, in the generation process of the traditional model component, a more experienced engineer is required to perform parameter adjustment and the like, and in the scheme, the threshold of the technology is greatly reduced, and a user can obtain the model component only by providing a target material set. Thus being very convenient for the popularization and the use of the audio frequency identification technology.
Drawings
In order to more clearly illustrate the embodiments of the present description or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present description, and that other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
Fig. 1 is a schematic diagram of a conference system according to an embodiment of the present disclosure;
FIG. 2 is a schematic flow chart of a method for generating model components according to an embodiment of the present disclosure;
fig. 3 is a schematic flow chart of a method for screening materials according to an embodiment of the present disclosure;
FIG. 4 is a schematic flow chart of a method for creating constraints according to an embodiment of the present disclosure;
FIG. 5 is a schematic flow chart of a method for creating constraint conditions according to an embodiment of the present disclosure;
FIG. 6 is a schematic flow chart of a method for creating constraints according to an embodiment of the present disclosure;
FIG. 7 is a schematic flow chart of a method for generating model components according to an embodiment of the present disclosure;
FIG. 8 is a schematic block diagram of a model component generating apparatus according to an embodiment of the present disclosure;
FIG. 9 is a schematic flow chart of a model component calling method according to an embodiment of the present disclosure;
FIG. 10 is a schematic flow diagram of a model component invocation method provided in an embodiment of the present disclosure;
FIG. 11 is a schematic block diagram of a model component calling device according to an embodiment of the present disclosure;
FIG. 12 is a schematic flow chart of a model component call method according to an embodiment of the present disclosure;
FIG. 13 is a schematic block diagram of a model component calling device according to an embodiment of the present disclosure;
FIG. 14 is a schematic flow chart of a model component call method according to an embodiment of the present disclosure;
fig. 15 is a schematic block diagram of a model component calling device according to an embodiment of the present disclosure.
Detailed Description
In order to make the technical solution in the present specification better understood by those skilled in the art, the technical solution in the present specification embodiment will be clearly and completely described with reference to the drawings in the present specification embodiment, and it is apparent that the described embodiment is only a part of the embodiment of the present specification, not all the embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.
Referring to fig. 1, in one example scenario, a company is provided with a conference system, which may include a server, a terminal device, and a client. The terminal device may be configured to generate audio data that is received by the client to generate an audio file. The server runs an audio recognition model component to recognize audio files recorded and provided by the terminal equipment and generate conference records. The server may be implemented using cloud computing technology. A company is a real estate enterprise, so that the communication content of multiple meetings is relevant to the real estate field.
In this scenario example, a user may send a request to a server through a client to cause a machine learning model component in the server to automatically generate a corresponding model component according to a user-specified requirement. In this scenario example, a server may have a machine learning model component therein that may generate an audio recognition model component as further described by way of example.
In this scenario example, the user may be ready for text material regarding the real estate field, e.g., legal regulations regarding the housing and the land trade, industry standards for the housing, contracts for the real estate industry, and so forth. The user may provide the text material to the server via the client. The machine learning model component of the server may automatically generate a final audio recognition model component from the provided material set.
In this scenario example, after receiving the text material, the server may perform operations such as word segmentation on the text material to obtain a material set in terms. For example, the material set may be { cell, floor height, rebar, construction, helmet, 8756732847, land unit price, floor price, property, 24845, rebar, plastic, construction, concrete, floor price, rebar, … … }.
In this scenario example, the server may filter the materials in the material set to obtain a target material set. Specifically, the server selects materials with certain semantics in the material set, so that numbers '8756732847' and '24845' which are not expressed by semantics are eliminated in the material set. And forming a target material set by taking the rest materials as target materials. At this time, the target material set may include { district, floor height, reinforcement, construction, helmet, land unit price, floor price, property, reinforcement, plastic, construction, concrete, floor price, reinforcement, … … }.
In this scenario example, the server may analyze and count the number of target stories, and further calculate a duty ratio of the number of each target story to the total amount in the target story set. For example, the number of "cells" is 13, the number of "layer heights" is 34, the number of "reinforcements" is 73, and so on, which are not listed. Assuming that the total amount of the materials in the target material set is 500, the number of the "cells" is 2.6%, the ratio of the "layer height" is 6.8%, and the ratio of the "steel bars" is 14.6. The different duty ratios can represent the frequency of the corresponding target materials in daily use to a certain extent. In the process of generating the audio recognition model component, the possibility of hit when matching with audio information of a relatively large target material can be correspondingly improved.
In this scenario example, there may be multiple standard model components in the server, for which the machine learning component may train or parameter modify, etc., to generate a customized audio recognition model component from the standard model components. The server may be formulated with a first constraint for constraining the generation of the audio recognition model component. Specifically, the first constraint condition may be a target material with a ratio greater than 10%, and when the audio information is matched with the target material, the weight corresponding to the target material is increased by 0.1. In the process of recognizing the audio information into characters, the audio recognition model component firstly characterizes the semantic information as a voice feature vector, and the target material is also characterized as a material feature vector. Further, the target material corresponding to the voice feature vector can be determined by calculating the matching degree between the voice feature vector and the material feature vector. For example, a maximum match approach may be taken. In this scenario example, the client formulates a first constraint such that the audio recognition model component also performs the first constraint in performing the vector matching. Specifically, when matching the audio feature vector with the material feature vector of the "steel bar", the calculated matching degree is assumed to be 0.3, and the weight 0.1 is further increased based on the matching degree of 0.3, that is, the matching degree of the final audio feature vector and the material feature vector of the "steel bar" is considered to be 0.4. Therefore, under the condition that the matching degree of one audio feature vector and a plurality of material feature vectors is relatively close, for example, the matching degree of 'rigid' is 0.31, the matching degree of 'dry' is 0.3, and the matching degree of 'reinforcing steel bar' is increased to 0.4, so that the hit probability of 'reinforcing steel bar' is increased. The frequency of the steel bar used in the field is high through the analysis of the target material set, so that the hit probability of the steel bar is increased, and the recognition accuracy of the audio recognition model component is improved.
In this scenario example, the standard model components in the server may each be generated based on a variety of algorithms. For example, the server may have standard model components that are generated separately for audio recognition according to hidden Markov, neural networks, etc. algorithms. Specifically, the server can automatically construct a quasi audio recognition model component according to different standard model components according to the target material set. For example, a first quasi audio recognition model component is generated according to a standard model component construction of a hidden Markov as a core algorithm. And constructing and generating a second quasi audio recognition model component according to the standard model component taking the convolutional neural network as a core algorithm.
In this scenario example, the generated quasi audio recognition model components may be filtered to obtain the final used audio recognition model components. Specifically, the first quasi audio recognition model component and the second quasi audio recognition model component can be trained based on the same number of target materials, and recognition accuracy of the first quasi audio recognition model component and the second quasi audio recognition model component is obtained. Namely, calculating the ratio of the number of correct recognition modules to the total amount of the first quasi audio recognition module according to the target materials, and calculating the ratio of the number of correct recognition modules to the total amount of the second quasi audio recognition module. Assuming that the recognition accuracy of the first quasi audio recognition model component is 99.8%, and the recognition accuracy of the second quasi audio recognition model component is 99.7%, the first quasi audio recognition model component can be used as a final audio recognition model component.
In this scenario example, after the server generates the audio recognition model component, the generated audio recognition model component may be further published. An audio recognition model component may be originally run in the server. After the machine learning model component generates the new audio recognition model component, the new audio recognition model component may be placed into a set of recognition model components that are used to store the audio recognition model component. In this manner, version management and storage of the audio recognition model components is facilitated. Further, the server may put the newly generated audio recognition model component directly into use to update the audio recognition model component that the server originally operates.
In this scenario example, after the audio recognition model component in the server is updated, the server may perform audio recognition on the audio file recorded by the terminal device to obtain a text file, thereby bringing convenience to the user. Because the audio recognition model component in the server is generated for automatically analyzing the material set, the whole updating process is convenient and quick. Furthermore, the target material set has clear content, so that the obtained audio recognition model component can have better recognition accuracy aiming at the target materials in the target material set.
Please refer to fig. 9. The embodiment of the specification provides a calling method of a model component. The calling method can be applied to a client, a server and a system formed by the client and the server. The calling method may include the following steps.
Step S62: providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component.
In this embodiment, the machine learning model component may be used to automatically generate the model component from the material set. The generated model component can realize certain business functions. The business functions may be constrained by the material set such that the model component is associated with the material set input to the machine learning model component. Further, the generated model components may be specified by differences in material sets.
In this embodiment, the machine learning model component may modify the specified model component to yield a customized model component. The specified model component may be a model component specified by a user, or may be a machine learning model component having a standard model component in advance. The model component specified by the user can be a model component which is invoked by the user before. After the user uses the model component, the model component is considered to have certain defects, so that the model component is specified, and the machine learning model component can be improved to a certain degree for the model component to obtain the customized model component. Furthermore, standard model components can be preset in the school sub-model components. The number of standard model components may be one or more. The standard models may be generated gradually based on the respective algorithms. For example, the standard model components may be generated based on a neural network, hidden Markov, etc., respectively, or the standard model components may be generated using a plurality of algorithms.
In this embodiment, the machine learning model component may automatically generate a customized model component from the specified model component based on analyzing the material set. In particular, for example, a user may not need to have a great knowledge of the entire build process of the customized model component. The machine learning model component can count the number of materials in the material set, and when the number is larger than a specified threshold, the specified model component can be a standard model component based on a neural network algorithm; otherwise, the specified model component may be a selection of standard model components based on a hidden Markov algorithm.
In this embodiment, the customized model component generated by the machine learning model component may be generated by training a specified model component based on a set of stories provided by the user. Because the customized model is generated by training the materials in the material set in the process of generating the customized model, the customized model component can respond to the materials in the material set better. Furthermore, the materials in the material set can be selected to be not accurate enough in processing by using the model component in the prior art, so that the generated customized model component can be improved in processing accuracy compared with the model component in the prior art.
In this embodiment, providing the material set to the machine learning model component may include: the user stores the material set in a specified path of the client and provides the path for the machine learning model component; or the client sends the material set to the server through the network, and the server receives the material set and then is further responded and processed by the machine learning model component.
In this embodiment, specifying a service function may refer to processing data or the like to meet a service requirement, in order to pursue a result required for the service requirement. The business requirement can be a condition that people need to achieve when they reach a certain purpose in production and life. In particular, for example, the specified business functions may include, but are not limited to, an audio recognition function, an image sharpening processing function, a video compression function, and so forth.
Step S64: and calling the customized model component to realize the specified business function.
In this embodiment, after the customized model component is generated, the customized model component may be replaced with the designated model component. Alternatively, the customized model component may be used in superposition with the specified model component. For example, the final result is output after the result output by the specified model component and the result output by the customized model component are evaluated in combination. For example, when performing text recognition in an image, the text in the image may be "first" in the world, and the result of recognition of the specified model component may be "first", "dry", "first", and thus, it may be known that the specified model component is not accurate enough to recognize "next" in "first" in the world, and thus, an image containing "first" in the world, or other images containing "next" may be used as material to generate the customized model component. So that the result of the customizable model component output for the above image may be "first in the world". Designating a model component as identifying "down" in an image may score a number of similar words, e.g., 0.5 for "dry" 0.4 for "down" and 0.2 for "big". When the customized model component recognizes "lower" in the image, the score of the similar word may be scored, for example, "dry" is 0.1, "lower" is 0.5, "big" is 0.1, and thus, after the scores of the same word are added, the word corresponding to the maximum value is selected as the output result, and the score of "lower" is 0.9 and is the maximum value. Therefore, the appointed model component and the customized model component are overlapped and used, and the recognition accuracy can be improved. Therefore, the customized model component can be used as a supplementary upgrading module of the specified model component to make up for the defect of the specified model component.
In this embodiment, invoking the customized model component may include: directly calling the appointed model component according to the calling address of the customized model component; or sending a call request to a specified address, wherein the call request is attached with service data, and the service data is used for being input to the customized model component to realize the service function.
Please refer to fig. 10. In one embodiment, the step of providing a set of stories to a machine learning model component for use by the machine learning model component in generating a customized model component from the set of stories may include the following steps.
Step S70: and according to the material set, a first constraint condition of the appointed model component is formulated, wherein the first constraint condition is used for constraining the training process of the appointed model component.
Step S72: and taking the material set as input, and generating the customized model component based on a training process of the designated model component according to the first constraint condition constraint.
In this embodiment, the machine learning component may take the material set as input and constrain a training process of the specified model component according to the first constraint condition to generate the customized model component. The first constraint may function to constrain the generation process of the customized model component. Generating the customized model component requires some process. Such as specifying algorithms, setting parameters, model training, etc. In this embodiment, constraint conditions may be formulated according to the material set, and the machine learning model component may select the specified model component according to the reading conditions. In the training process of the appointed model component, parameters and the like of the appointed model component are set through constraint conditions, so that the finally generated customized model component can meet the requirements. Specifically, for example, audio recognition with respect to the knowledge field to which the material set corresponds is satisfied. In the present embodiment, the number of the first constraint conditions is not limited, and may be one or more, for example.
Please refer to fig. 3. In one embodiment, when providing a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component, the following steps may be included.
Step S30: receiving a material set input by a user; wherein the material set includes a plurality of materials.
Step S32: screening the material set according to a preset rule to obtain a target material; and the screened target materials form a target material set.
Step S34: the target material set is provided to the machine learning model component for the machine learning model component to generate the customized model component based on the specified model component in accordance with the target material set.
In this embodiment, the target material may be obtained by screening the material provided by the user. Thus, the target material for training the model component can have better data consistency. In this embodiment, the material set provided by the user may be text material that may appear in the case of the primary usage scenario of the user. Specifically, for example, the model component is applied to a court for generating a court trial record for a recording of a court trial. At this time, the text in the trial court record may be used as the target material.
In this embodiment, the preset rules may include, but are not limited to: selecting the expressed content as the material in the appointed field; or selecting the material with the data format as the specified format. Specifically, the specified domain may refer to a certain business domain or knowledge domain. For example, the designated area may be football area, basketball area, civil law area, aviation area, loan area, etc., which are not listed. In this embodiment, the content of the expression is a material in a specified area, and may be words, phrases, pictures, or videos that are frequently used in the specified area. Specifically, for example, "three-ball" is a common word in the field of basketball; "wing" is a common term in the aeronautical field. The specified format may be a data format specifying the target material. Specifically, for example, if the specified format is text format, then the text format material is taken as the target material, and the non-text format pictures and videos are not put into the target material set.
Please refer to fig. 11. The embodiment of the specification also provides a calling device of the model component, which comprises: a generation module that provides a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component; and the calling module is used for calling the customized model component to realize the specified service function.
The functions and effects implemented by the modules in the calling device provided in this embodiment may be explained in reference to other embodiments, and are not described herein again.
The present description also provides a computer storage medium storing computer program instructions that, when executed, implement: providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component; and calling the customized model component to realize the specified business function.
In the present embodiment, the computer storage medium includes, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read-Only Memory (ROM), a Cache (Cache), a Hard Disk (HDD), or a Memory Card (Memory Card).
The functions and effects of the computer storage medium provided in this embodiment when the program instructions are executed may be explained by referring to other embodiments, and are not described herein.
Please refer to fig. 12. The embodiment of the specification provides a calling method of a model component. The method can be applied to clients. The client may be an electronic device with certain data processing capabilities, such as a desktop computer, a notebook computer, a tablet computer, a smart phone, or the like. Of course, the client may also refer to software running in the electronic device. The method may comprise the following steps.
Step S82: receiving a material set designated by a user; wherein the material set includes at least one material.
Step S84: transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component.
Step S86: and receiving the customized model component fed back by the server.
Step S88: and calling the customized model component to realize the specified service function.
In this embodiment, the customized model component may be run in a client. The user may operate the client to send the set of specified stories to the server to cause the machine learning model component of the server to generate the customized model component. Further, after the machine learning model component generates the customized model component, the customized model component may be sent to the client. In this way, the client may invoke the customized model component to effect an update to the local model component. The process is very simple and convenient, and brings convenience to users.
Please refer to fig. 13. The embodiment of the specification also provides a calling device of the model component, which may include: the first receiving module is used for receiving a material set appointed by a user; wherein the material set comprises at least one material; the sending module is used for sending the material set to a server, so that a machine learning model component of the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component; the second receiving module is used for receiving the customized model component fed back by the server; and the calling module is used for calling the customized model component to realize the specified service function.
The functions and effects implemented by the modules in the calling device provided in this embodiment may be explained in reference to other embodiments, and are not described herein again.
The present description provides a computer storage medium storing computer program instructions that, when executed, implement: receiving a material set designated by a user; wherein the material set comprises at least one material; transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified model component is used as a basis for generating the customized model component; receiving a customized model component fed back by the server; and calling the customized model component to realize the specified service function.
The functions and effects of the computer storage medium provided in this embodiment when the program instructions are executed may be explained by referring to other embodiments, and are not described herein.
Please refer to fig. 14. The embodiment of the specification also provides a calling method of the model component. The calling method may include the following steps.
Step S92: the set of stories is sent to a server for the server to take the set of stories as input to a machine learning model component and generate a customized model component based on the specified model component.
Step S94: and sending a call request to call the customized model component to realize the specified service function.
In this embodiment, the calling method of the model component may run on the client. The client may have relatively weak data processing capabilities. In this way, the customized model component can run in the server, and the client can call the customized model component by sending a corresponding call request to the server.
According to the method and the device, the operation capability requirement on the electronic equipment applying the calling method can be reduced, and therefore the use and popularization of the model component can be facilitated.
Please refer to fig. 15. The embodiment of the specification also provides a calling device of the model component. The calling means of the model component may include: the system comprises a material sending module, a machine learning model component, a customization model component and a material processing module, wherein the material sending module is used for sending a material set to a server, and the server is used for taking the material set as the input of the machine learning model component and generating the customization model component based on the specified model component; and the request sending module is used for sending a calling request to call the customized model component to realize the specified service function.
The functions and effects implemented by the modules in the calling device provided in this embodiment may be explained in reference to other embodiments, and are not described herein again.
The present description also provides a computer storage medium having stored therein computer program instructions that, when executed, implement: transmitting the material set to a server for the server to take the material set as input to a machine learning model component and generate a customized model component based on a specified model component; and sending a call request to call the customized model component to realize the specified service function.
The functions and effects of the computer storage medium provided in this embodiment when the program instructions are executed may be explained by referring to other embodiments, and are not described herein.
Please refer to fig. 2. The embodiment of the specification provides a model component generating method. The model component generating method operates in the electronic device. The electronic device may be a server or a terminal device with better performance, such as a desktop computer or a notebook computer. The model component generating method may include the following steps.
Step S20: determining a material set for generating a customized model component; wherein the material set includes a plurality of materials.
In this embodiment, the custom model component may be a functional module that is programmed by a computer to implement a certain business function. Specifically, for example, the customized model component may include: an audio recognition model component for recognizing audio data from which text data can be recognized; or an image processing model component for processing the image to enhance the sharpness of the image; or a video processing model component for enhancing the image clarity of a video data presentation, etc. The customized model component may be a customized model component that the user automatically generates to meet the business requirements based on the machine learning model component of the server providing the business requirements.
In this embodiment, the material may be a data object for training a specified model component as input in the customized model component generation process. The material may include: text, images, audio or video, etc. The material may be about a certain knowledge domain. The model component obtained in this way can have better processing accuracy to the materials in the field. In particular, for example, the generated model components are used in the field of knowledge related to football, and the materials may include text materials, video files, images and the like related to football event news, football match video, football star pictures, football newspaper reviews, football periodicals, ball star profiles and the like. Of course, the method is not limited to the football field, but can be a music art field, a legal field and the like, and is not repeated herein.
In one embodiment, the material may be text. Specifically, the material may be words, words or phrases. Text material may be disaggregated by setting a word segmentation machine or the like. For example, the input text is "car running at high speed", which can be split into "high speed", "traveling", "running" and "car" by a word splitter having a semantic analysis function, or into "high speed" and "car", or can directly use "car running at high speed" as a material.
In this embodiment, the material set may be a data set formed of a plurality of materials. Specifically, the materials can be formed into a material set by adopting data structures such as a data table, an array and the like.
In this embodiment, the manner of determining the material set may include: receiving a material set provided by a user; or, according to the knowledge field appointed by the user, automatically acquiring the material of the knowledge field to form a material set; or receiving the material set sent by the client.
Step S22: and according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining the training process of the designated model component.
In this embodiment, the model component generating method may automatically further generate a model component according to the material set after determining the material set. Therefore, manual control and adjustment are not needed, and labor is saved. Furthermore, in the conventional process of generating model components, a highly experienced expert is required to control and adjust the generating process, such as setting parameters and the like. In the embodiment, through program setting, the material set can be automatically identified, and further constraint conditions for generating the customized model component are formulated. Therefore, the workload of generating the customized model component can be greatly reduced, and the popularization and application of the technology are facilitated.
In this embodiment, the first constraint is used to constrain the training process of the specified model component. The training process of the specified model component may include parameter setting, training speed control, setting of initial values, etc. for the specified model component. In this embodiment, the training process of the specified model component is restricted by the constraint condition, so that the finally generated customized model component can meet the requirements. The number of the first constraint conditions is not limited, and may be one or more, for example.
In this embodiment, constraint conditions are formulated according to the material set, so that a user can provide materials, and a customized model component can be further obtained. No significant knowledge of the user about the entire build process of the model component may be required. Formulating constraints from the material set may include, but is not limited to: counting the number of materials in the material set, and designating an algorithm for generating a model component as a neural network when the number is larger than a designated threshold value, otherwise, designating the algorithm for generating the model component as a hidden Markov; or calculating the duty ratio of each material in the material set, and formulating constraint conditions according to the size of the duty ratio to improve the weight of the material with larger duty ratio when matching with the audio information so as to improve the hit rate of the material; or, in the case of analyzing knowledge fields related to the material set and involving more than two knowledge fields, constraint conditions are formulated to prescribe that at least more than two algorithms are adopted to respectively generate a model component. Of course, those skilled in the art may make other modifications and variations, and the functions and effects of the present invention are the same as or similar to those of the present application, and all such modifications and variations are intended to be included in the scope of the present application.
Step S24: and taking the material set as input, and obtaining the customized model component based on a training process of the specified model component according to the first constraint condition constraint.
In this embodiment, in the process of generating the customized model component, the material set is used as training data for specifying the model component. The customized model component finally obtained in this way can have better recognition accuracy to the material set in the material set. Further, the training process is constrained by the first constraint condition, so that the whole process is relatively targeted. Specifically, for example, the first constraint specifies that the algorithm is a convolutional neural network, and the number of layers is 3. At this time, the specified model component may be a standard model component adopting a convolutional neural network and has a 3-layer network structure, and features of the audio information are extracted, so that a speech feature vector formed by the audio information is matched with feature vectors of the materials, and a final identified result is obtained.
In the embodiment of the specification, constraint conditions are formulated by analyzing the material set so as to constrain the generation process of the customized model component. Therefore, after the material set is specified, the model component can be automatically generated without further manual intervention. The manual labor is reduced. In addition, in the generation process of the traditional model component, a more experienced engineer is required to perform parameter adjustment and the like, and in the scheme, the threshold of the technology is greatly reduced, and a user can obtain the model component only by providing a material set. Thus being very convenient for the popularization and the use of the audio frequency identification technology.
Please refer to fig. 4. In one embodiment, the first constraint condition of the specified model component is formulated according to the material set, and the following steps are included.
Step S36: and calculating the duty ratio of the materials in the material set.
Step S38: formulating the first constraint includes: and adjusting the weight of the materials corresponding to the duty ratio in the appointed model component when matching is carried out according to the duty ratio.
In this embodiment, the number of each material in the material set may be calculated, and then the ratio of the number of each material to the total amount may be calculated. Specifically, for example, there are 1000 materials in a material set, wherein the materials "computers" are 50 in total, and the material may have a 50/1000 ratio.
In this embodiment, the duty ratio of the material may be expressed to some extent, and the frequency of use of the material may be expressed. The occupied area is relatively large, and the use frequency of the material is relatively high; the occupied area is smaller, and the use frequency of the material is relatively smaller.
In this embodiment, during the working process of the model component, the feature vector is generally used to characterize the service data, and the material is characterized by the material feature vector. The business data may be data that is input required to execute the model component. And determining whether the material characterized by the material feature vector is selected or not through the matching degree of the feature vector and the material feature vector. Weights can be configured for the matching degree of different material feature vectors so as to correct deviation brought by a matching algorithm. Further, since the duty ratio may be used to reflect the frequency of use of the corresponding material, the weight of the material having a relatively high duty ratio may be relatively increased, or the weight of the material having a relatively low duty ratio may be relatively decreased. Specifically, for example, the service data may be audio information, and the above manner may improve recognition accuracy of text recognition on the audio data.
Please refer to fig. 5. In one embodiment, the first constraint condition of the specified model component is formulated according to the material set, and the following steps are included.
Step S40: and counting the quantity of the materials in the material set.
Step S42: formulating the first constraint includes: an algorithm for specifying the use of the model component is determined based on the number.
In this embodiment, the specified model component may be generated based on a variety of algorithms. Specifically, for example, algorithms that may be used to generate the audio recognition model component include, but are not limited to: hidden markov or neural network algorithms, etc. Alternatively, algorithms for generating image recognition model components include, but are not limited to: deep learning, reinforcement learning, and the like. Different algorithms, there may be some difference for different number levels, etc. For example, most algorithms can have better training results and operation speed in the case of smaller amounts of material. However, if the number of materials is very large, some algorithms cause a very large amount of computation due to the large number of materials, and a large computation load is imposed on the electronic device. Thus, formulating the first constraint includes further specifying an algorithm based on the number. Therefore, the more adaptive algorithm can be automatically selected according to the actual situation of the material set.
In this embodiment, the determination of the algorithm used to generate the specified model component may be implemented in a manner that the standard model component is selected by the algorithm. The standard model component may be generated based on an algorithm. In this manner, by determining the algorithm that generated the model component, a standard model component can be specified for further adoption by the machine learning model component.
In this embodiment, the data table may be set, or the program instructions may be integrated. After the number of materials in the material set is counted, a specified algorithm and a specified model component are further obtained.
Please refer to fig. 6. In one embodiment, the step of formulating the first constraint of the model component based on the material set may include the following steps.
Step S44: dividing the knowledge field to which the materials in the material set belong.
Step S46: formulating the first constraint includes: in the case where the material set involves more than two knowledge domains, at least more than two specified model components are specified.
In this embodiment, some algorithms have better recognition accuracy for the generated customized model component for the content of a certain knowledge domain, but the recognition accuracy of the generated customized model component may not be high for the rest of knowledge domains. In the present embodiment, when the material in the material set relates to two or more knowledge domains, a plurality of specified model components are specified to correspond to the knowledge domains, respectively. Thus, after the plurality of specified model components complete the training process, a plurality of quasi-customized model components are obtained. The method can select among the generated multiple quasi-customized model components, and takes the quasi-customized model components which can relatively better consider all knowledge fields as the customized model components for final use. Specifically, for example, the material set relates to the stock domain and the ship domain, and the formulated first constraint condition may be to select two specified model components and be constructed based on a long-short-term memory algorithm and a convolutional neural network algorithm, respectively.
In one embodiment, the method for generating the customized model component may further include: receiving a second constraint condition input by a user; wherein the second constraint is used to constrain a training process of the specified model component.
Correspondingly, when the customized model component is generated, the material set is used as input, and the customized model component is generated based on the training process of the specified model component according to the first constraint condition and the second constraint condition.
In this embodiment, the second constraint condition may be specified by a user input method, in addition to the first constraint condition formulated by analysis based on the material set. The second constraint may also constrain the training process of the specified model component. When the first constraint condition and the second constraint condition conflict, the second constraint condition may be determined. Of course, the first constraint may also be satisfied. Specifically, for example, the first constraint condition is to specify and select a specified model component generated according to the hidden markov algorithm, and the second constraint condition is to select a specified model component generated according to the neural network algorithm, where the second constraint condition input by the default user may be executed in preference to the first constraint condition, and the specified model component constructed by the neural network algorithm is adopted. Of course, the first constraint and the second constraint may also be implemented separately, such as selecting two specified model components, one of which is generated according to the first constraint and the other of which is generated according to the second constraint.
In this embodiment, when the first constraint condition and the second constraint condition do not contradict, the first constraint condition and the second constraint condition jointly constrain the generation process of the model component. Specifically, for example, a first constraint is to generate two quasi-customized model components, and a second constraint is to specify that the algorithm is hidden Markov. At this time, two quasi-customized model components may be generated using a specified model component based on the hidden markov algorithm, or one quasi-customized model component may be generated using a specified model component based on the hidden markov algorithm, and another quasi-customized model component may be generated using a specified model component of another algorithm. And one of the generated more than two quasi-customized model components can be selected as a final customized model component.
Please refer to fig. 7. In one embodiment, when the material set is used as an input, the model component is generated according to the generation process of the first constraint condition constraint, and the method may include the following steps.
Step S50: and taking the material set as input, and generating more than two quasi-customized model components based on the training process of the specified model components according to the first constraint condition constraint.
Step S52: and evaluating more than two quasi-customized model components to determine the customized model components.
In this embodiment, two or more quasi-customized model components may be generated without violating the first constraint. The quasi-customized model components can be respectively generated by training according to preset specified model components. The quasi-customized model component is still constrained by the first constraint during training. Specifically, for example, the first constraint condition is that a neural network algorithm is adopted, and the plurality of quasi-customized model components can respectively generate quasi-customized model components by adopting specified model components based on a convolutional neural network, a cyclic neural network, a deep neural network and the like.
In this embodiment, the number of customized model components specified for the first constraint condition may be used to constrain the number of customized model components that are output finally, and the number of aligned customized model components in the intermediate generation process is not limited. Therefore, the final model component can be selected from the generated multiple quasi-customized model components, and the quasi-customized model component with better performance and better fit for the use requirement is selected as the final model component.
In this embodiment, the step of evaluating the two or more quasi-customized model components and determining the model components includes: and calculating the identification accuracy of each quasi-customized model component, and taking the identification accuracy as the final customized model component.
In this embodiment, when the customized model component is aligned and selected, the customized model component that is ultimately used can be better applied to the use environment according to the recognition accuracy of the quasi-customized model component. The recognition accuracy can be a ratio of the number of accurate recognition to the number of inaccurate recognition or a ratio of the number of accurate recognition to the total amount after a certain number of materials are recognized by the calculation quasi-customized model component in the training process.
In this embodiment, the final customized model component with high recognition accuracy may be understood as selecting the quasi-customized model component with the highest recognition accuracy as the final customized model component. The final customized model component can also be determined for comprehensively considering the recognition accuracy, the operation feedback speed, the operation workload and the like. The final customized model component is not necessarily the quasi-customized model component with the highest recognition accuracy.
In one embodiment, the model component generating method may further include: publishing the customized model component to a model component set for obtaining the customized model component from the model component set; wherein the set of model components includes at least one customized model component.
In this embodiment, the model component set may be a set formed by a plurality of customized model components. The generated model components are put into the model component set for storage, so that the version management of the model components can be facilitated. When a specific model component needs to be used online, the corresponding model component can be pushed to a corresponding business unit. Of course, the service unit may also issue an acquisition request to acquire the corresponding model component.
In this embodiment, after the model component is generated and placed in the model component set, the model component may be automatically sent to the corresponding service unit. Therefore, the method and the device can automatically release and use the latest model components without manual operation, and bring convenience to operation users.
Please refer to fig. 8. The embodiment of the specification also provides a model component generating device, which comprises: the system comprises a determining module, a formulating module and a generating module.
A determining module for determining a material set for generating a customized model component; wherein the material set comprises a plurality of materials;
the formulating module is used for formulating a first constraint condition of the appointed model component according to the material set, wherein the first constraint condition is used for constraining the training process of the appointed model component;
and the generation module is used for taking the material set as input and generating the customized model component based on the training process of the specified model component according to the first constraint condition constraint.
In this embodiment, the related content may be explained in comparison with the content of the foregoing embodiment, and will not be described herein.
The present description also provides a computer storage medium storing computer program instructions that, when executed, implement: determining a set of materials for generating a customized model component, wherein the set of materials includes a plurality of materials; according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining a training process of the designated model component; and taking the material set as input, and obtaining the customized model component based on a training process of the specified model component according to the first constraint condition constraint.
In the present embodiment, the computer storage medium includes, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read-Only Memory (ROM), a Cache (Cache), a Hard Disk (HDD), or a Memory Card (Memory Card).
The functions and effects of the computer storage medium provided in this embodiment when the program instructions are executed may be explained by referring to other embodiments, and are not described herein.
The various embodiments in the specification are described in a progressive manner, and identical and similar parts of the various embodiments are mutually referred to, and each embodiment is mainly described as a difference from other embodiments.
The server mentioned in the embodiment of the present specification may be an electronic device having a certain arithmetic processing capability. Which may have network communication terminals, processors, memory, and the like. Of course, the server may also refer to software running in the electronic device. The server may be a distributed server, or may be a system having a plurality of processors, memories, network communication modules, etc. that cooperate. The server can also be realized by adopting a cloud computing technology.
In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable Gate Array, FPGA)) is an integrated circuit whose logic function is determined by the programming of the device by a user. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented with "logic compiler" software, which is similar to the software compiler used in program development and writing, and the original code before the compiling is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but HDL is not only one, but a plurality of kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware DescriptionLanguage), confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), lava, lola, myHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog2 are most commonly used at present. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.
Those skilled in the art will also appreciate that, in addition to implementing clients, servers in the form of pure computer readable program code, it is well possible to implement the same functions by logically programming method steps such that clients, servers are implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, and the like. Such clients, servers may therefore be considered as a hardware component, and the means included therein for performing various functions may also be considered as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.
From the above description of embodiments, it will be apparent to those skilled in the art that the present description may be implemented in software plus a necessary general purpose hardware platform. Based on this understanding, the technical solution of the present specification may be embodied in essence or a part contributing to the prior art in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the method described in the various embodiments or some parts of the embodiments of the present specification.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are referred to each other, and each embodiment is mainly described as different from other embodiments. In particular, for the embodiments of the apparatus, the computer storage medium, reference may be made to the description of the embodiments of the method described above for comparison.
The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
Although the present description has been described by way of embodiments, those of ordinary skill in the art will recognize that there are many variations and modifications to the present description without departing from the spirit of the present description, and it is intended that the appended claims encompass such variations and modifications without departing from the spirit of the present description.

Claims (28)

1. A method for invoking a model component, comprising:
providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
and calling the customized model component to realize the specified business function.
2. The method of claim 1, wherein the specified model component comprises at least one of: a model component specified by a user, or a standard model component preset by the machine learning model component.
3. The method of claim 1, wherein in the step of providing a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component comprises:
according to the material set, a first constraint condition of the appointed model component is formulated, wherein the first constraint condition is used for constraining a training process of the appointed model component;
and taking the material set as input, and obtaining the customized model component based on a training process of the specified model component according to the first constraint condition constraint.
4. A method according to claim 3, wherein the step of formulating the first constraint of the specified model component based on the set of materials comprises:
calculating the duty ratio of each material in the material set;
formulating the first constraint includes: adjusting the weight corresponding to the material when matching the material with the audio information in the appointed model component according to the duty ratio; the weight is used for correcting the calculated deviation of the matching degree of the materials and the audio information.
5. A method as claimed in claim 3, wherein, in the step of formulating a first constraint for the specified model component from the material set, it comprises:
counting the quantity of the materials in the material set;
formulating the first constraint includes: an algorithm for specifying the use of the model component is determined based on the number.
6. A method as claimed in claim 3, wherein the step of formulating the first constraint of the specified model component from the set of materials comprises:
dividing the knowledge field to which the materials in the material set belong;
formulating the first constraint includes: in the case where the material set involves more than two knowledge domains, at least two or more specified model components are specified for the corresponding knowledge domain.
7. The method of claim 1, wherein in the step of providing a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component comprises:
receiving a material set input by a user; wherein the material set comprises a plurality of materials;
screening the material set according to a preset rule to obtain a target material; the target materials obtained through screening form the target material set;
The target material set is provided to the machine learning model component for the machine learning model component to generate the customized model component based on the specified model component in accordance with the target material set.
8. The method of claim 7, wherein the preset rules include at least one of: selecting the expressed content as a material in a designated field; or selecting the material with the data format as the specified format.
9. A method as claimed in claim 3, wherein the method further comprises:
receiving a second constraint condition input by a user; wherein the second constraint is used to constrain a training process of the specified model component;
accordingly, in the step of generating the customized model component,
and taking the material set as input, and generating the customized model component based on the training process of the designated model component according to the first constraint condition and the second constraint condition.
10. The method of claim 3, wherein in the step of generating the customized model component based on the training process in which the specified model component is constrained according to the first constraint, taking the material set as input, comprising:
Taking the material set as input, and generating more than two quasi-customized model components according to the generation process of the first constraint condition constraint;
and evaluating more than two quasi-customized model components to determine the customized model components.
11. The method of claim 10, wherein in evaluating for more than two of the quasi-customized model components, determining the customized model components comprises:
and calculating the recognition rate of each quasi-customized model component, and taking the recognition rate which is higher as the final customized model component.
12. A model component recall device, comprising:
a generation module that provides a set of stories to a machine learning model component for the machine learning model component to generate a customized model component from the set of stories and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
And the calling module is used for calling the customized model component to realize the specified service function.
13. A computer storage medium having stored thereon computer program instructions that when executed perform: providing the material set to a machine learning model component for the machine learning model component to generate a customized model component from the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
and calling the customized model component to realize the specified business function.
14. A method for invoking a model component, comprising:
Receiving a material set designated by a user; wherein the material set comprises at least one material;
transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
receiving a customized model component fed back by the server;
and calling the customized model component to realize the specified service function.
15. A model component recall device, comprising:
the first receiving module is used for receiving a material set appointed by a user; wherein the material set comprises at least one material;
The sending module is used for sending the material set to a server, so that a machine learning model component of the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
the second receiving module is used for receiving the customized model component fed back by the server;
and the calling module is used for calling the customized model component to realize the specified service function.
16. A computer storage medium storing computer program instructions that when executed implement: receiving a material set designated by a user; wherein the material set comprises at least one material; transmitting the material set to a server, so that a machine learning model component for the server generates a customized model component according to the material set and based on a specified model component; wherein the material set comprises at least one material; the customized model component is used for realizing specified service functions; the specified business function comprises at least one of voice recognition and image recognition realized based on a material set; the generating a customized model component according to the material set and based on a specified model component comprises the following steps: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component; receiving a customized model component fed back by the server; and calling the customized model component to realize the specified service function.
17. A method for invoking a model component, comprising:
transmitting the material set to a server for the server to take the material set as input to a machine learning model component and generate a customized model component based on a specified model component; the generating a customized model component based on the specified model component includes: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
sending a call request to call the customized model component to realize a specified service function; the specified business function comprises at least one of voice recognition and image recognition realized based on the material set.
18. A model component recall device, comprising:
the system comprises a material sending module, a machine learning model component, a customization model component and a material processing module, wherein the material sending module is used for sending a material set to a server, and the server is used for taking the material set as the input of the machine learning model component and generating the customization model component based on the specified model component; the generating a customized model component based on the specified model component includes: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component;
The request sending module is used for sending a calling request to call the customized model component to realize the specified service function; the specified business function comprises at least one of voice recognition and image recognition realized based on the material set.
19. A method of generating a model component, comprising:
determining a material set for generating a customized model component; wherein the material set comprises a plurality of materials;
according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining a training process of the designated model component;
taking the material set as input, and obtaining the customized model component based on the training process of the specified model component according to the first constraint condition constraint; wherein, include: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component; the customized model component is used for realizing the function of making business; the specified business function comprises at least one of voice recognition and image recognition realized based on the material set.
20. The method of claim 19, wherein, in the step of formulating a first constraint specifying a model component from the set of stories, comprising:
calculating the duty ratio of the materials in the material set;
formulating the first constraint includes: and adjusting the weight of the materials corresponding to the duty ratio in the appointed model component when matching is carried out according to the duty ratio.
21. The method of claim 19, wherein, in the step of formulating a first constraint specifying a model component from the set of stories, comprising:
counting the quantity of the materials in the material set;
formulating the first constraint includes: an algorithm for specifying the use of the model component is determined based on the number.
22. The method of claim 19, wherein, in the step of formulating the first constraint of the specified model component from the material set, comprising:
dividing the knowledge field to which the materials in the material set belong;
formulating the first constraint includes: in the case where the material set involves more than two knowledge domains, at least two or more specified model components are specified for the corresponding knowledge domain.
23. The method of claim 19, wherein the method further comprises:
Receiving a second constraint condition input by a user; wherein the second constraint is used to constrain a training process of the specified model component;
accordingly, in the step of generating the customized model component,
and taking the material set as input, and generating the customized model component based on the training process of the designated model component according to the first constraint condition and the second constraint condition.
24. The method of claim 19, wherein in the step of generating the model component based on the generation of the specified model component in accordance with the first constraint condition constraint taking the material set as input comprises:
taking the material set as input, and generating more than two quasi-customized model components based on the training process of the specified model components according to the first constraint condition constraint;
and evaluating more than two quasi-customized model components to determine the customized model components.
25. The method of claim 24, wherein in evaluating for more than two of the quasi-customized model components, determining the customized model components comprises:
And calculating the recognition rate of each quasi-customized model component, and taking the recognition rate which is higher as the final customized model component.
26. The method of claim 19, wherein the method further comprises:
publishing the customized model component to a model component set for obtaining the customized model component from the model component set; wherein the set of model components includes at least one customized model component.
27. A model component generating apparatus, comprising:
a determining module for determining a material set for generating a customized model component; wherein the material set comprises a plurality of materials;
the formulating module is used for formulating a first constraint condition of the appointed model component according to the material set, wherein the first constraint condition is used for constraining the training process of the appointed model component;
the generation module is used for taking the material set as input and generating the customized model component based on the training process of the specified model component according to the first constraint condition constraint; wherein, include: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component; the customized model component is used for realizing the function of making business; the specified business function comprises at least one of voice recognition and image recognition realized based on the material set.
28. A computer storage medium storing computer program instructions that when executed implement: determining a set of materials for generating a customized model component, wherein the set of materials includes a plurality of materials; according to the material set, a first constraint condition of a designated model component is formulated, wherein the first constraint condition is used for constraining a training process of the designated model component; taking the material set as input, and obtaining the customized model component based on the training process of the specified model component according to the first constraint condition constraint; wherein, include: calculating the duty ratio of each material in the material set; determining the weight corresponding to the material according to the duty ratio; the weights are used to correct the likelihood that the target material is hit when speech recognition and/or image recognition is implemented using the customized model component; the customized model component is used for realizing the function of making business; the specified business function comprises at least one of voice recognition and image recognition realized based on the material set.
CN201810769756.0A 2018-07-13 2018-07-13 Model component calling and generating method, device and storage medium Active CN110716767B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810769756.0A CN110716767B (en) 2018-07-13 2018-07-13 Model component calling and generating method, device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810769756.0A CN110716767B (en) 2018-07-13 2018-07-13 Model component calling and generating method, device and storage medium

Publications (2)

Publication Number Publication Date
CN110716767A CN110716767A (en) 2020-01-21
CN110716767B true CN110716767B (en) 2023-05-05

Family

ID=69208501

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810769756.0A Active CN110716767B (en) 2018-07-13 2018-07-13 Model component calling and generating method, device and storage medium

Country Status (1)

Country Link
CN (1) CN110716767B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112184498A (en) * 2020-09-29 2021-01-05 中国平安财产保险股份有限公司 Contract scoring method and device, computer equipment and storage medium
CN114579110B (en) * 2022-05-05 2022-08-19 支付宝(杭州)信息技术有限公司 Solution method of optimization model, electronic device, application program, and storage medium
CN116302294B (en) * 2023-05-18 2023-09-01 安元科技股份有限公司 Method and system for automatically identifying component attribute through interface

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1719454A (en) * 2005-07-15 2006-01-11 清华大学 Off-line hand writing Chinese character segmentation method with compromised geomotric cast and sematic discrimination cost
CN103914702A (en) * 2013-01-02 2014-07-09 国际商业机器公司 System and method for boosting object detection performance in videos
CN105448292A (en) * 2014-08-19 2016-03-30 北京羽扇智信息科技有限公司 Scene-based real-time voice recognition system and method
CN107644638A (en) * 2017-10-17 2018-01-30 北京智能管家科技有限公司 Audio recognition method, device, terminal and computer-readable recording medium
CN107807814A (en) * 2017-09-27 2018-03-16 百度在线网络技术(北京)有限公司 Construction method, device, equipment and the computer-readable recording medium of application component
CN108229519A (en) * 2017-02-17 2018-06-29 北京市商汤科技开发有限公司 The method, apparatus and system of image classification

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9070360B2 (en) * 2009-12-10 2015-06-30 Microsoft Technology Licensing, Llc Confidence calibration in automatic speech recognition systems
TWI634441B (en) * 2016-11-29 2018-09-01 財團法人工業技術研究院 Method to enhance association rules, apparatus using the same and computer readable medium

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1719454A (en) * 2005-07-15 2006-01-11 清华大学 Off-line hand writing Chinese character segmentation method with compromised geomotric cast and sematic discrimination cost
CN103914702A (en) * 2013-01-02 2014-07-09 国际商业机器公司 System and method for boosting object detection performance in videos
CN105448292A (en) * 2014-08-19 2016-03-30 北京羽扇智信息科技有限公司 Scene-based real-time voice recognition system and method
CN108229519A (en) * 2017-02-17 2018-06-29 北京市商汤科技开发有限公司 The method, apparatus and system of image classification
CN107807814A (en) * 2017-09-27 2018-03-16 百度在线网络技术(北京)有限公司 Construction method, device, equipment and the computer-readable recording medium of application component
CN107644638A (en) * 2017-10-17 2018-01-30 北京智能管家科技有限公司 Audio recognition method, device, terminal and computer-readable recording medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Multi-domain learning by confidence-weighted parameter combination;Mark Dredze等;《Machine Learning》;123–149 *
几种小训练样本集的数字语音识别模型的比较性研究;贺苏宁;虞厥邦;;计算机科学(09);全文 *

Also Published As

Publication number Publication date
CN110716767A (en) 2020-01-21

Similar Documents

Publication Publication Date Title
CN111460150A (en) Training method, classification method and device of classification model and storage medium
CN110716767B (en) Model component calling and generating method, device and storage medium
CN112215171B (en) Target detection method, device, equipment and computer readable storage medium
CN111626049A (en) Title correction method and device for multimedia information, electronic equipment and storage medium
US11275994B2 (en) Unstructured key definitions for optimal performance
CN111258995A (en) Data processing method, device, storage medium and equipment
US20220383854A1 (en) Intent recognition method and intent recognition system having self learning capability
US20230368028A1 (en) Automated machine learning pre-trained model selector
CN111783873A (en) Incremental naive Bayes model-based user portrait method and device
CN114118192A (en) Training method, prediction method, device and storage medium of user prediction model
US11636282B2 (en) Machine learned historically accurate temporal classification of objects
CN112202849A (en) Content distribution method, content distribution device, electronic equipment and computer-readable storage medium
CN111507726A (en) Message generation method, device and equipment
CN113569017B (en) Model processing method and device, electronic equipment and storage medium
US20230141408A1 (en) Utilizing machine learning and natural language generation models to generate a digitized dynamic client solution
CN111783453B (en) Text emotion information processing method and device
CN112799658B (en) Model training method, model training platform, electronic device, and storage medium
CN115129902A (en) Media data processing method, device, equipment and storage medium
CN114491093A (en) Multimedia resource recommendation and object representation network generation method and device
CN114417944B (en) Recognition model training method and device, and user abnormal behavior recognition method and device
CN114817526B (en) Text classification method and device, storage medium and terminal
CN114328797B (en) Content search method, device, electronic apparatus, storage medium, and program product
US11526544B2 (en) System for object identification
CN117291168A (en) Aspect emotion analysis method and system
CN117453273A (en) Intelligent program code complement method and device

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
REG Reference to a national code

Ref country code: HK

Ref legal event code: DE

Ref document number: 40021951

Country of ref document: HK

GR01 Patent grant
GR01 Patent grant