CN111459842A - Mobile phone APP automatic test method based on N L P and KG - Google Patents
Mobile phone APP automatic test method based on N L P and KG Download PDFInfo
- Publication number
- CN111459842A CN111459842A CN202010443059.3A CN202010443059A CN111459842A CN 111459842 A CN111459842 A CN 111459842A CN 202010443059 A CN202010443059 A CN 202010443059A CN 111459842 A CN111459842 A CN 111459842A
- Authority
- CN
- China
- Prior art keywords
- app
- mobile phone
- node
- layer
- path
- 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.)
- Pending
Links
- 238000010998 test method Methods 0.000 title claims abstract description 8
- 238000012360 testing method Methods 0.000 claims abstract description 54
- 238000004364 calculation method Methods 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 19
- 239000013598 vector Substances 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 7
- 238000003491 array Methods 0.000 claims description 5
- 230000002457 bidirectional effect Effects 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 abstract description 35
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000013499 data model Methods 0.000 abstract description 4
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000010276 construction Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013522 software testing Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3688—Test management for test execution, e.g. scheduling of test suites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3692—Test management for test results analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3696—Methods or tools to render software testable
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/31—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M1/00—Substation equipment, e.g. for use by subscribers
- H04M1/24—Arrangements for testing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Computer Hardware Design (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses an automatic test method of a mobile phone APP based on N L P and KG, which relates to the technical field of software test, wherein the method utilizes a DSSM semantic similarity calculation technology, a BERT natural language reasoning technology and a knowledge graph storage technology to automatically construct each function of an application program into a graph data structure, each path in the graph represents a complete operation flow for completing a certain function of the APP, compared with the traditional automatic test, the mobile phone APP is completely tested off-line on the basis of a trained deep learning model, and the test case can be reused by automatically generating a natural language script and a knowledge graph data model of the mobile phone APP.
Description
Technical Field
The invention relates to the technical field of software testing, in particular to an automatic android mobile phone APP testing method based on N L P and KG.
Background
Natural language processing is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that enable efficient communication between humans and computers using natural language. Is the field of computer science, artificial intelligence, linguistics focusing on the interaction between computers and human (natural) language.
With the wide popularization of 4G and 5G networks and the continuous improvement of hardware technology, the types and the number of application programs suitable for smart phones are continuously increased, but as the page design in the software becomes more complex and the functions are continuously increased, more software faults and problems are caused, which brings huge challenges to the software testing work. In mobile phone software design company, because of the closure of mobile phone software platform design, the original software black box test is basically realized through manual test, in recent years, the difficulty of manual test is higher and higher due to the appearance of smart phones, the product design period is continuously shortened, and the design cost is continuously compressed, so that each company starts to try to replace manual work with automatic test, the test cost is reduced, and the product quality is improved.
In the automatic test system, because a large number of operations interacting with the mobile phone APP exist and basic functions of the mobile phone APP in a characteristic field, such as a bank APP, are the same, the operation rules of the mobile phone APP in the same field can be learned by using natural language processing N L P and a deep learning technology, and an operation graph of the mobile phone APP can be constructed by using a knowledge graph data model.
Disclosure of Invention
The invention aims to provide an automatic testing method of a mobile phone APP based on N L P and KG aiming at a traditional automatic testing system of APP functionality, which is used for improving the testing speed and precision of the APP.
The technical scheme is as follows:
an automatic test method for a mobile phone APP based on N L P and KG comprises the following specific steps:
firstly, uploading data (such as login accounts and the like) required by a mobile phone APP test, and connecting a mobile phone and an APP for testing;
calling a general path generation module to generate general paths of the APP, such as login and registration, and storing the general paths into a general path list;
step (3) extracting the general path from the general path list, sequentially extracting each operation node or scene node in the current path, and simultaneously acquiring APP page information corresponding to the current node through a page information extraction function;
step (4) calling a DSSM model, matching to obtain the APP page control name and the space operation script corresponding to the current node, executing the next page challenged by the current script, and repeatedly executing the steps 3 and 4;
and (5) if the nodes in the path are extracted, calling a BERT model to carry out operation process reasoning on the last node in the path and all control names of the current APP page, generating a new path for the control which is judged to be the next operation of the current node by combining the previous operation process, adding the new path into the current path list, and then executing the script content corresponding to the new node. Repeatedly executing the 5 th operation;
step (6), when the node type is a result node, namely representing that the current operation flow is executed completely, the current operation flow can be stored in a result list, and the step 3 is executed;
and (7) after all the operation paths are supported and generated, converting the operation paths into a preset knowledge graph data structure through a test graph generation function, storing the preset knowledge graph data structure into a graph database, and simultaneously producing a test report for a tester to check.
In a further embodiment, the semantic similarity calculation model DSSM may be divided into three layers from bottom to top: the system comprises an input layer, a presentation layer and a matching layer, wherein the input layer, the presentation layer and the matching layer are used for realizing the search of a general path and the matching of a node and the content of a page control in the automatic test process of a mobile phone APP page by using a DSSM-based semantic similarity calculation model; the input layer of the DSSM converts the two input sentences into two-dimensional arrays represented by Word vectors (vectors trained by CBOW models based on negative sampling in Word2 Vec) respectively, and the two-dimensional arrays are used as the input of the representation layer;
the method comprises the following steps of coding position information of words in a sentence by a DSSM (direct sequence markup language) representation layer by using a bidirectional GRU (generalized regression) model, coding the meaning information by using a Transformer coder part to replace a traditional M L P structure, converting sentence feature representation originally consisting of each word vector into new sentence representation by global-attribute, and setting w i as a feature vector of the ith word and N as the length of the sentence, wherein the calculation formula of the global-attribute is as follows:
ki＝tanh(Wswi+bs),i∈[0，N]
after the feature vectors of two sentences are obtained by the representation layer, the semantic similarity between the two sentences can be represented by the cosine similarity and cosine distance of the two feature vectors, and the calculation formula of the cosine similarity is as follows:
in the invention, N represents a node, and W represents page characters.
In a further embodiment, the BERT model is a multi-layer bidirectional Transformer encoder composed of a plurality of small encoders, wherein each small encoder is composed of a self-attribute layer and a full-connect layer, respectively, and is realized by using a BERT-based natural language reasoning technology to perform an operation flow reasoning function in the automatic test process of the mobile phone APP.
In a further embodiment, the input sentences A and B in the BERT-based N L I model are spliced together for joint modeling, and the two sentences are separated by a sentence separator [ SEP ], and a classification identifier [ C L S ] is introduced into the head of the spliced sentences, wherein the classification identifier is mainly used for classifying the inference result.
[CLS]A[SEP]B
Has the advantages that:
compared with the manual function test of the APP, the function test method has the obvious advantages that the function test speed of the APP is improved, the defect that manual test is easy to fatigue and make mistakes is overcome, the test efficiency is improved, the popularization of the mobile phone application program to the market is accelerated, and more profits are created.
Compared with the traditional automatic test system, the invention has the obvious advantages that the mobile phone APP is completely tested off line on the basis of the trained deep learning model, and the test cases can be reused by automatically generating the natural language script and the knowledge graph data model of the mobile phone APP.
Drawings
FIG. 1 is a functional block diagram of the system of the present invention.
FIG. 2 is an APP automatic test flow diagram of the present invention.
Fig. 3 shows a structure diagram of a DSSM model.
FIG. 4 shows a diagram of the transform encoder architecture.
Fig. 5 shows a layer complete structure.
FIG. 6 shows a global-orientation structure
N L I model structure of the BERT of FIG. 7
Detailed Description
Compared with the traditional automatic test system, on the basis of the deep learning model after training, the mobile phone APP carries out off-line test completely, and the test cases can be reused by automatically generating the natural language script and the knowledge graph data model of the mobile phone APP.
As can be seen from fig. 1, the whole system is divided into five subsystems, namely a page information extraction subsystem, a semantic similarity calculation subsystem, an operation flow reasoning subsystem, an operation flow automatic generation subsystem and a test chart subsystem; the system comprises twelve functions, namely a page information acquisition function, a control script generation function, a semantic similarity calculation model training, managing and calling function, an operation flow reasoning model training, managing and calling function, an operation flow generation function, a common sense library construction function, a test map construction function and a test map storage function.
(1) The page information extraction subsystem: the page information extraction function mainly acquires control information on a corresponding page of the APP by interacting with the APP, and according to the acquired page information, the following two subfunctions are completed: a. a page information formatting function, b. a control script automatic generation function;
(2) the semantic similarity calculation operator system mainly uses a semantic similarity calculation model DSSM of N L P, pages such as a login page and a registration page are searched in the automatic APP test process, and then the functions of login, registration and the like are automatically completed according to the searched corresponding pages.
(3) And the operation flow reasoning subsystem mainly uses an operation flow reasoning model BERT of N L P to automatically deduce whether the operation B of the APP is the next step of the operation A or not according to the operation rule of the APP in the corresponding field, thereby helping the APP to generate a complete operation flow.
(4) The operation flow automatically generates a subsystem: by calling a page information extraction function, a semantic similarity calculation function and an operation flow reasoning function, a complete operation flow of each function (such as a transfer function in a bank APP) of the APP is automatically generated.
(5) And the test map generation subsystem automatically generates a test map structure predefined in advance according to the complete operation flow of each function of the mobile phone APP generated by the operation flow automatic generation function, and stores the test map structure in the map database for later-stage test personnel to reuse.
In order to realize the functions, the system is provided with thirteen modules which are respectively a control information acquisition module, a control script generation module, a semantic similarity calculation model training, managing and calling module, an operation flow reasoning model training, managing and calling module, a general path generation module, a complete path generation module, a common sense library construction module, a test map construction module and a test map storage module.
(1) A control information acquisition module: the method is responsible for interacting with the APP to be tested in the Android mobile phone in the automatic test process, continuously acquiring information of all controls in the corresponding page of the APP, and processing and converting the format of the obtained page control data.
(2) The control script generation module: and the system is responsible for receiving the preprocessed control data and generating a natural language execution script of the corresponding control according to the APP data (account number, password and the like) to be tested input by the WEB terminal tester, so that the natural language execution script is used by the operation flow generation subsystem.
(3) A problem matching module: if the user feeds back the correct answer, the subsequent operation is stopped, the whole question-answering process is finished, and if the user feeds back the wrong answer, the question-answering of the type is selected from the knowledge base according to the question type selected by the user to carry out semantic similarity calculation on the data.
(4) The semantic similarity calculation model training, managing and calling module comprises: and by calling the pre-trained DSSM model, similarity calculation is carried out on the obtained APP page, so that related pages such as login and registration are found, and the functions such as login and registration are completed.
(5) The operation flow reasoning model training, managing and calling module comprises: whether the operation B of the APP is the next step of the operation A or not is automatically deduced by calling a BERT model which finishes training for the APP in a certain field according to the operation rule of the APP in the corresponding field, so that the APP is assisted to generate a complete operation flow.
(6) A general path generation module: and the system is responsible for automatically generating an initial stage in the operation process, continuously interacting with the APP to be tested, and generating corresponding login and registration paths according to login/registration related data provided by WEB testers.
(7) A complete path generation module: and the system is responsible for continuously interacting with the APP to be tested according to the login/registration path obtained by the general path generation module and the related data provided by WEB terminal testers, and continuously generating an operation path corresponding to the subsequent function after the execution of the login/registration operation of the APP is completed.
(8) A test map construction module: the system is responsible for reading a complete APP operation flow stored by the automatic generation subsystem of the operation flow, converting the complete APP operation flow into a data structure required by the knowledge graph and counting related data, so that the test graph storage module can store conveniently and the counting result is returned to WEB testers.
(9) The test map storage module: and repeating the duty according to the graph database address, the account password and the like provided by the WEB terminal tester, and storing the test graph after the data format conversion into the Neo4j data.
As can be seen from fig. 2, a specific processing flow of the method for automatically testing the APP of the mobile phone based on N L P and KG is as follows:
firstly, uploading data (such as login accounts and the like) required by a mobile phone APP test, and connecting a mobile phone and an APP for testing;
calling a general path generation module to generate general paths of the APP, such as login and registration, and storing the general paths into a general path list;
step (3) extracting the general path from the general path list, sequentially extracting each operation node or scene node in the current path, and simultaneously acquiring APP page information corresponding to the current node through a page information extraction function;
step (4) calling a DSSM model, matching to obtain the APP page control name and the space operation script corresponding to the current node, executing the next page challenged by the current script, and repeatedly executing the steps 3 and 4;
and (5) if the nodes in the path are extracted, calling a BERT model to carry out operation process reasoning on the last node in the path and all control names of the current APP page, generating a new path for the control which is judged to be the next operation of the current node by combining the previous operation process, adding the new path into the current path list, and then executing the script content corresponding to the new node. Repeatedly executing the 5 th operation;
step (6), when the node type is a result node, namely representing that the current operation flow is executed completely, the current operation flow can be stored in a result list, and the step 3 is executed;
and (7) after all the operation paths are supported and generated, converting the operation paths into a preset knowledge graph data structure through a test graph generation function, storing the preset knowledge graph data structure into a graph database, and simultaneously producing a test report for a tester to check.
Fig. 3 to 6 are structural diagrams of a model for calculating semantic similarity, which are introduced as follows:
the DSSM may be divided into three layers from bottom to top: the structure of the input layer, the presentation layer, and the matching layer is shown in FIG. 3 below.
1. Input layer
The input layer of the DSSM converts the two sentences input into two-dimensional arrays represented by Word vectors (vectors trained using the negative-sampling-based CBOW model in Word2 Vec) respectively, and takes them as the input of the representation layer.
2. Presentation layer
The DSSM representation layer firstly encodes the position information of words in a sentence by using a bidirectional GRU model, then encodes the semantic information by using a Transformer encoder part (as shown in figure 4) to replace the traditional M L P structure, and finally converts the sentence characteristic representation originally consisting of each word vector into a new sentence representation through global-attributeiThe feature vector of the ith word is shown, and N is the sentence length, the calculation formula of global-attribute is as follows:
ki＝tanh(Wswi+bs)，i∈[0，N]
the complete structure of the DSSM-representative layer and the global-attribute calculation structure are shown in fig. 5 and 6, respectively.
3. Matching layer
After the feature vectors of two sentences are obtained by the representation layer, the semantic similarity between the two sentences can be represented by the cosine similarity of the two feature vectors, namely the cosine distance, and the calculation formula of the cosine similarity is as follows:
in the automatic test method, N represents a node, and W represents a page text.
FIG. 7 is a block diagram of the BERT model used to perform operational flow reasoning, the model of which is described below:
the transform-based bi-directional coded representation (BERT) model structure is a multi-layer bi-directional transform Encoder. For the encoder, it consists of a plurality of small encoders, each of which consists of a self-attention layer and a full-connect layer, respectively, as shown in fig. 4.
For the BERT-based N L I model, the structure is shown in FIG. 7, unlike the CNN + ATTENTION model, which requires the computation of the feature representations of sentences A and B, respectively, for which the input sentences A and B are spliced together for joint modeling and separated by a sentence separator [ SEP ], a classification identifier [ C L S ] is introduced at the head of the spliced sentences, the classification identifier is mainly used for the classification of the inference result, and finally, the input of the whole model is in the form of [ C L S ] A { SEP ] B.
Claims (4)
1. An automatic test method for a mobile phone APP based on N L P and KG is characterized by comprising the following specific steps:
firstly, uploading data (such as login accounts and the like) required by a mobile phone APP test, and connecting a mobile phone and an APP for testing;
calling a general path generation module to generate general paths of the APP, such as login and registration, and storing the general paths into a general path list;
step (3) extracting the general path from the general path list, sequentially extracting each operation node or scene node in the current path, and simultaneously acquiring APP page information corresponding to the current node through a page information extraction function;
step (4) calling a DSSM model, matching to obtain the APP page control name and the space operation script corresponding to the current node, executing the next page challenged by the current script, and repeatedly executing the steps 3 and 4;
and (5) if the nodes in the path are extracted, calling a BERT model to carry out operation process reasoning on the last node in the path and all control names of the current APP page, generating a new path for the control which is judged to be the next operation of the current node by combining the previous operation process, adding the new path into the current path list, and then executing the script content corresponding to the new node. Repeatedly executing the 5 th operation;
step (6), when the node type is a result node, namely representing that the current operation flow is executed completely, the current operation flow can be stored in a result list, and the step 3 is executed;
and (7) after all the operation paths are supported and generated, converting the operation paths into a preset knowledge graph data structure through a test graph generation function, storing the preset knowledge graph data structure into a graph database, and simultaneously producing a test report for a tester to check.
2. The automatic test method for the mobile phone APP based on the N L P and the KG as claimed in claim 1, wherein the semantic similarity calculation model DSSM can be divided into three layers from bottom to top, including an input layer, a presentation layer and a matching layer, wherein the input layer of the DSSM converts two input sentences into two-dimensional arrays represented by Word vectors (vectors trained by CBOW model based on negative sampling in Word2 Vec) respectively and takes the two-dimensional arrays as the input of the presentation layer;
the DSSM representation layer uses a bidirectional GRU model to encode the position information of words in a sentence, uses a Transformer encoder part to replace the traditional M L P structure to encode the semantic information, converts the sentence feature representation originally composed of each word vector into a new sentence representation through global-attribute, and sets wiThe feature vector of the ith word is shown, and N is the sentence length, the calculation formula of global-attribute is as follows:
ki＝tanh(Wswi+bs)，i∈[0，N]
after the feature vectors of two sentences are obtained by the representation layer, the semantic similarity between the two sentences can be represented by the cosine similarity and cosine distance of the two feature vectors, and the calculation formula of the cosine similarity is as follows:
in the invention, N represents a node, and W represents page characters.
3. The method for automatically testing mobile phone APP based on N L P and KG of claim 1, wherein the BERT model is a multi-layer bidirectional transform encoder consisting of a plurality of small encoders, wherein each small encoder consists of a self-attribute layer and a full-connect layer.
4. The method for automatically testing mobile phone APP based on N L P and KG as claimed in claim 1, wherein the input sentences A and B in the BERT-based N L I model are spliced together for joint modeling, and the two sentences are separated by a sentence separator [ SEP ], and a classification identifier [ C L S ] is introduced into the head of the spliced sentence, and the classification identifier is mainly used for classification of the inference result.
[CLS]A [SEP]B。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010443059.3A CN111459842A (en) | 2020-05-22 | 2020-05-22 | Mobile phone APP automatic test method based on N L P and KG |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010443059.3A CN111459842A (en) | 2020-05-22 | 2020-05-22 | Mobile phone APP automatic test method based on N L P and KG |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111459842A true CN111459842A (en) | 2020-07-28 |
Family
ID=71678805
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010443059.3A Pending CN111459842A (en) | 2020-05-22 | 2020-05-22 | Mobile phone APP automatic test method based on N L P and KG |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111459842A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160321321A1 (en) * | 2013-09-06 | 2016-11-03 | Microsoft Technology Licensing, Llc | Deep structured semantic model produced using click-through data |
| CN110502643A (en) * | 2019-08-28 | 2019-11-26 | 南京璇玑信息技术有限公司 | A kind of next model autocreating technology of the prediction based on BERT model |
| CN110516240A (en) * | 2019-08-28 | 2019-11-29 | 南京璇玑信息技术有限公司 | A kind of Semantic Similarity Measurement model DSSM technology based on Transformer |
| CN111078546A (en) * | 2019-12-05 | 2020-04-28 | 北京云聚智慧科技有限公司 | Method for expressing page features and electronic equipment |
-
2020
- 2020-05-22 CN CN202010443059.3A patent/CN111459842A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160321321A1 (en) * | 2013-09-06 | 2016-11-03 | Microsoft Technology Licensing, Llc | Deep structured semantic model produced using click-through data |
| CN110502643A (en) * | 2019-08-28 | 2019-11-26 | 南京璇玑信息技术有限公司 | A kind of next model autocreating technology of the prediction based on BERT model |
| CN110516240A (en) * | 2019-08-28 | 2019-11-29 | 南京璇玑信息技术有限公司 | A kind of Semantic Similarity Measurement model DSSM technology based on Transformer |
| CN111078546A (en) * | 2019-12-05 | 2020-04-28 | 北京云聚智慧科技有限公司 | Method for expressing page features and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 盐味橙汁: "《详解注意力机制（Attention）-Global/Local/Self Attetion》", 《HTTPS://BLOG.CSDN.NET/JASONZHOUJX/ARTICLE/DETAILS/83386627》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114116994A (en) | Welcome robot dialogue method | |
| CN109241330A (en) | The method, apparatus, equipment and medium of key phrase in audio for identification | |
| CN109857846B (en) | Method and device for matching user question and knowledge point | |
| CN108763535A (en) | Information acquisition method and device | |
| CN112989796B (en) | Text naming entity information identification method based on syntactic guidance | |
| CN114722839B (en) | Man-machine cooperative dialogue interaction system and method | |
| CN111930906A (en) | Knowledge graph question-answering method and device based on semantic block | |
| CN116596347B (en) | Multi-disciplinary interaction teaching system and teaching method based on cloud platform | |
| CN108959388A (en) | information generating method and device | |
| CN112527986A (en) | Multi-round dialog text generation method, device, equipment and storage medium | |
| CN113704419A (en) | Conversation processing method and device | |
| CN113268610A (en) | Intent skipping method, device and equipment based on knowledge graph and storage medium | |
| CA3135717A1 (en) | System and method for transferable natural language interface | |
| CN113392265A (en) | Multimedia processing method, device and equipment | |
| CN116166688A (en) | Business data retrieval method, system and processing equipment based on natural language interaction | |
| CN110516240B (en) | Semantic similarity calculation model DSSM (direct sequence spread spectrum) technology based on Transformer | |
| CN113326367B (en) | Task type dialogue method and system based on end-to-end text generation | |
| Hafeth et al. | Semantic representations with attention networks for boosting image captioning | |
| CN111414466A (en) | Multi-round dialogue modeling method based on depth model fusion | |
| CN111459842A (en) | Mobile phone APP automatic test method based on N L P and KG | |
| CN115935969A (en) | Heterogeneous data feature extraction method based on multi-mode information fusion | |
| CN114625759A (en) | Model training method, intelligent question answering method, device, medium, and program product | |
| CN113010676B (en) | Text knowledge extraction method, device and natural language inference system | |
| CN113010662B (en) | Hierarchical conversational machine reading understanding system and method | |
| CN115130461A (en) | Text matching method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200728 |