CN116611799B - Cloud office-based data processing method, AI cloud office server and medium - Google Patents
Cloud office-based data processing method, AI cloud office server and medium Download PDFInfo
- Publication number
- CN116611799B CN116611799B CN202310903597.XA CN202310903597A CN116611799B CN 116611799 B CN116611799 B CN 116611799B CN 202310903597 A CN202310903597 A CN 202310903597A CN 116611799 B CN116611799 B CN 116611799B
- Authority
- CN
- China
- Prior art keywords
- moment
- text
- text semantic
- semantic
- session
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 280
- 238000001914 filtration Methods 0.000 claims abstract description 128
- 238000000034 method Methods 0.000 claims abstract description 39
- 230000006835 compression Effects 0.000 claims abstract description 37
- 238000007906 compression Methods 0.000 claims abstract description 37
- 238000004458 analytical method Methods 0.000 claims abstract description 30
- 230000008569 process Effects 0.000 claims abstract description 9
- 238000003860 storage Methods 0.000 claims abstract description 8
- 238000009795 derivation Methods 0.000 claims description 27
- 230000007704 transition Effects 0.000 claims description 21
- 230000010354 integration Effects 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 4
- 230000002776 aggregation Effects 0.000 description 4
- 238000004220 aggregation Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000004931 aggregating effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer And Data Communications (AREA)
Abstract
According to the cloud office based data processing method, the AI cloud office server and the medium, the text semantic moment with the initial high text fine granularity and the text semantic moment obtained by the feature compression can be mixed in detail, the problem of detail loss caused by the feature compression is avoided, in addition, a text feature processing component formed by a plurality of feature processing cores is adopted in the sliding window filtering process, procedural information generated in the sliding window filtering process is not required to be reserved, and further, on the basis of ensuring the cloud office scene session analysis quality, the occupation of storage resources of the AI cloud office server in the cloud office scene session processing process can be avoided as much as possible.
Description
Technical Field
The invention relates to the technical field of cloud office, in particular to a data processing method based on cloud office, an AI cloud office server and a medium.
Background
Cloud office refers to the establishment of office on the basis of cloud computing technology, so that office cost is reduced, office efficiency is improved, and low-carbon emission reduction is realized. Advantages of cloud office include, compared to traditional office: the compatibility of office software is improved, simultaneous editing and modification of a plurality of persons of a document are realized, a network virtual knowledge production group is built at any time, cloud office data is synchronized and accessed at any time and any place through a mobile internet, and the like. Along with the continuous popularization of cloud office, the analysis value of the cloud office data information is not ignored, and the traditional technology is mostly based on artificial intelligence to realize the analysis processing of the cloud office data information, but the technology is difficult to ensure the analysis quality of the cloud office data information, and can occupy excessive storage resources.
Disclosure of Invention
The invention provides a data processing method based on cloud office, an AI cloud office server and a medium.
The invention provides a cloud office-based data processing method, which is applied to an AI cloud office server and comprises the following steps:
obtaining an initial text semantic moment and a processed text semantic moment corresponding to a current cloud office scene session, wherein the current cloud office scene session and the initial text semantic moment both correspond to a first text fine granularity, the processed text semantic moment corresponds to a second text fine granularity, and the second text fine granularity is smaller than the first text fine granularity;
according to the initial text semantic moment, a first text semantic moment is obtained through a first text feature processing component contained in an AI expert system network, wherein the first text semantic moment corresponds to the first text fine granularity, the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing;
obtaining a second text semantic moment through a second text feature processing component contained in the AI expert system network according to the processed text semantic moment, wherein the second text semantic moment corresponds to the second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection;
Obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after feature derivation operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after feature compression operation, wherein the third text semantic moment corresponds to the first text granularity and the fourth text semantic moment corresponds to the second text granularity;
and obtaining a session analysis result corresponding to the current cloud office scene session through the AI expert system network according to the third text semantic moment and the fourth text semantic moment.
In some examples, the obtaining a third text semantic moment using the first text semantic moment and the second text semantic moment after the feature derivation operation, obtaining a fourth text semantic moment using the second text semantic moment and the first text semantic moment after the feature compression operation, includes:
according to the first text semantic moment, a first linkage text semantic moment is obtained through a first sliding window filtering unit contained in the AI expert system network, wherein the first linkage text semantic moment corresponds to the first text fine granularity;
Performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity;
generating the third text semantic moment by using the first linkage text semantic moment and the second linkage text semantic moment;
performing feature compression operation on the first text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity;
obtaining a fourth linkage text semantic moment through a fourth sliding window filtering unit contained in the AI expert system network according to the second text semantic moment, wherein the fourth linkage text semantic moment corresponds to the second text fine granularity;
generating a fourth text semantic moment by using the third linkage text semantic moment and the fourth linkage text semantic moment;
or, the obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after the feature deriving operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after the feature compressing operation, including:
Performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity;
generating the third text semantic moment by using the first text semantic moment and the second linkage text semantic moment;
performing feature compression operation on the first text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity;
and generating the fourth text semantic moment by using the third linkage text semantic moment and the second text semantic moment.
In some examples, the obtaining, by a first text feature processing component included in the AI expert system network, a first text semantic moment according to the initial text semantic moment includes:
generating a first text semantic moment to be processed and a second text semantic moment to be processed by using the initial text semantic moment, wherein the sum of the description layer numbers of the first text semantic moment to be processed and the second text semantic moment to be processed is equal to the description layer number of the initial text semantic moment;
Performing multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through the first text feature processing component contained in the AI expert system network to obtain a first text convolution semantic moment;
performing multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through the first text feature processing component contained in the AI expert system network to obtain a second text convolution semantic moment;
generating the first text semantic moment by using the first text convolution semantic moment and the second text convolution semantic moment, wherein the number of description layers of the first text semantic moment is equal to the sum of the numbers of description layers of the first text convolution semantic moment and the second text convolution semantic moment;
the step of obtaining a second text semantic moment through a second text feature processing component included in the AI expert system network according to the processed text semantic moment includes:
generating a third to-be-processed text semantic moment and a fourth to-be-processed text semantic moment by using the processing text semantic moment, wherein the sum of the description layer numbers of the third to-be-processed text semantic moment and the fourth to-be-processed text semantic moment is equal to the description layer number of the processing text semantic moment;
Performing multi-round sliding window filtering processing on the third text semantic moment to be processed through the second text feature processing component contained in the AI expert system network to obtain a third text convolution semantic moment;
performing multi-round sliding window filtering processing on the fourth text semantic moment to be processed through the second text feature processing component contained in the AI expert system network to obtain a fourth text convolution semantic moment;
and generating the second text semantic moment by using the third text convolution semantic moment and the fourth text convolution semantic moment, wherein the number of description layers of the second text semantic moment is equal to the sum of the number of description layers of the third text convolution semantic moment and the fourth text convolution semantic moment.
In some examples, the performing, by the first text feature processing component included in the AI expert system network, multiple rounds of sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed to obtain a first text convolution semantic moment includes:
the first text feature processing component contained in the AI expert system network is used for carrying out least one-round sliding window filtering processing on the first text semantic moment to be processed to obtain a first transition text semantic moment;
The first transition text semantic moment is subjected to downsampling processing through an integral downsampling unit contained in the first text feature processing assembly, so that first text semantic knowledge is obtained;
processing the first text semantic knowledge through a knowledge integration unit contained in the first text feature processing component to obtain second text semantic knowledge, wherein the size of the second text semantic knowledge is larger than that of the first text semantic knowledge;
multiplying the first transition text semantic moment by adopting the second text semantic knowledge to obtain a second transition text semantic moment;
and obtaining the first text convolution semantic moment through the first text feature processing component contained in the AI expert system network according to the second transition text semantic moment and the second text semantic moment to be processed.
In some examples, the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment includes:
acquiring linked text semantic moments of the third text semantic moment through a fifth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the third text semantic moment correspond to the first text fine granularity;
Performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fourth text semantic moment, wherein the linked text semantic moment of the fourth text semantic moment corresponds to the first text fine granularity;
generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment;
generating a session disassembly result by utilizing the target text semantic moment corresponding to the current cloud office scene session;
and transmitting the session disassembly result to a session big data processing system so that the session big data processing system outputs the session disassembly result.
In some examples, the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment includes:
performing feature compression operation on the third text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the third text semantic moment, wherein the linked text semantic moment of the third text semantic moment corresponds to the second text fine granularity;
Acquiring linked text semantic moments of the fourth text semantic moment through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the fourth text semantic moment correspond to the second text fine granularity;
acquiring a first target text semantic moment through a ninth sliding window filtering unit contained in the AI expert system network according to the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment;
obtaining a second target text semantic moment through a downsampling unit contained in the AI expert system network according to the first target text semantic moment;
according to the second target text semantic moment, acquiring a risk discrimination possibility map corresponding to the current cloud office scene session through a knowledge integration unit contained in the AI expert system network;
determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using a risk discrimination possibility map;
and issuing the session risk discrimination viewpoint to a session big data processing system so that the session big data processing system outputs the session risk discrimination viewpoint.
In some examples, the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment includes:
obtaining a fifth text semantic moment through a third text feature processing component contained in the AI expert system network according to the third text semantic moment, wherein the fifth text semantic moment corresponds to the first text fine granularity, and the third text feature processing component comprises at least one feature processing core in cascade connection;
obtaining a sixth text semantic moment through a fourth text feature processing component contained in the AI expert system network according to the fourth text semantic moment, wherein the sixth text semantic moment corresponds to the second text fine granularity, and the fourth text feature processing component comprises at least one feature processing core in cascade connection;
and obtaining a session analysis result corresponding to the current cloud office scene session through the AI expert system network according to the fifth text semantic moment and the sixth text semantic moment.
In some examples, the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the fifth text semantic moment and the sixth text semantic moment includes:
Acquiring linked text semantic moments of the fifth text semantic moment through a fifth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the fifth text semantic moment correspond to the first text fine granularity;
performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the sixth text semantic moment, wherein the linked text semantic moment of the sixth text semantic moment corresponds to the first text fine granularity;
generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment;
generating a session disassembly result by utilizing the target text semantic moment corresponding to the current cloud office scene session;
and issuing the session risk discrimination viewpoint to a session big data processing system so that the session big data processing system outputs the session disassembly result.
In some examples, the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the fifth text semantic moment and the sixth text semantic moment includes:
Performing feature compression operation on the fifth text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fifth text semantic moment, wherein the linked text semantic moment of the fifth text semantic moment corresponds to the second text fine granularity;
acquiring linked text semantic moments of the sixth text semantic moment through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the sixth text semantic moment correspond to the second text fine granularity;
obtaining a first target text semantic moment through a ninth sliding window filtering unit contained in the AI expert system network according to the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment;
obtaining a second target text semantic moment through a downsampling unit contained in the AI expert system network according to the first target text semantic moment;
according to the second target text semantic moment, acquiring a risk discrimination possibility map corresponding to the current cloud office scene session through a knowledge integration unit contained in the AI expert system network;
Determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using a risk discrimination possibility map;
and issuing the session risk discrimination viewpoint to a session big data processing system so that the session big data processing system outputs the session risk discrimination viewpoint.
In some independent embodiments, the step of debugging the AI expert system network includes:
obtaining a sample cloud office scene session, wherein the sample cloud office scene session corresponds to an authenticated session disassembly result, and the authenticated session disassembly result is annotation information of the sample cloud office scene session on each text unit;
obtaining a sample initial text semantic moment and a sample processing text semantic moment corresponding to the sample cloud office scene session through an AI expert system network, wherein the sample initial text semantic moment and the sample cloud office scene session both correspond to a first text granularity, the sample processing text semantic moment corresponds to a second text granularity, and the second text granularity is smaller than the first text granularity;
obtaining a first text semantic moment through a first text feature processing component contained in an AI expert system network according to the sample initial text semantic moment, wherein the first text semantic moment corresponds to the first text fine granularity, the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing;
Obtaining a second sample text semantic moment through a second text feature processing component contained in the AI expert system network according to the sample processing text semantic moment, wherein the second sample text semantic moment corresponds to the second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection;
obtaining a third sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment after feature derivation operation, and obtaining a fourth sample text semantic moment by using the second sample text semantic moment and the first sample text semantic moment after feature compression operation, wherein the third sample text semantic moment corresponds to the first text granularity and the fourth sample text semantic moment corresponds to the second text granularity;
according to the third sample text semantic moment and the fourth sample text semantic moment, a session disassembly prediction result corresponding to the current cloud office scene session is obtained through the AI expert system network;
and optimizing the network variable of the AI expert system network through network debugging cost by utilizing the session disassembly prediction result and the authenticated session disassembly result.
In some independent embodiments, the step of debugging the AI expert system network includes:
obtaining a sample cloud office scene session, wherein the sample cloud office scene session corresponds to an authenticated session risk discrimination viewpoint, and the authenticated session risk discrimination viewpoint is information obtained after annotating a session risk subject of the sample cloud office scene session;
obtaining a sample initial text semantic moment and a sample processing text semantic moment corresponding to the sample cloud office scene session through an AI expert system network, wherein the sample initial text semantic moment and the sample cloud office scene session both correspond to a first text granularity, the sample processing text semantic moment corresponds to a second text granularity, and the second text granularity is smaller than the first text granularity;
obtaining a first text semantic moment through a first text feature processing component contained in an AI expert system network according to the sample initial text semantic moment, wherein the first text semantic moment corresponds to the first text fine granularity, the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing;
Obtaining a second sample text semantic moment through a second text feature processing component contained in the AI expert system network according to the sample processing text semantic moment, wherein the second sample text semantic moment corresponds to the second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection;
obtaining a third sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment after feature derivation operation, and obtaining a fourth sample text semantic moment by using the second sample text semantic moment and the first sample text semantic moment after feature compression operation, wherein the third sample text semantic moment corresponds to the first text granularity and the fourth sample text semantic moment corresponds to the second text granularity;
according to the third text semantic moment and the fourth text semantic moment, obtaining a session risk prediction viewpoint corresponding to the current cloud office scene session through the AI expert system network;
and optimizing the network variable of the AI expert system network through network debugging cost by utilizing the session risk prediction viewpoint and the authenticated session risk discrimination viewpoint.
The invention also provides an AI cloud office server, which comprises a processor and a memory; the processor is in communication with the memory, and the processor is configured to read and execute a computer program from the memory to implement the method described above.
The present invention also provides a computer readable storage medium having stored thereon a computer program which, when run, implements the method described above.
The technical scheme provided by the embodiment of the invention can comprise the following beneficial effects: firstly, an initial text semantic moment and a processing text semantic moment corresponding to a current cloud office scene session can be obtained, then according to the initial text semantic moment, a first text semantic moment is obtained through a first text feature processing component contained in an AI expert system network, then according to the processing text semantic moment, a second text semantic moment is obtained through a second text feature processing component contained in the AI expert system network, then a third text semantic moment is obtained through the first text semantic moment and the second text semantic moment after feature deriving operation, a fourth text semantic moment is obtained through the second text semantic moment and the first text semantic moment after feature compressing operation, and finally according to the third text semantic moment and the fourth text semantic moment, a session analysis result corresponding to the current cloud office scene session is obtained through the AI expert system network. By the design, the text semantic moment with the initial high text granularity and the text semantic moment obtained by the feature compression can be subjected to detail mixing, the problem of detail loss caused by the feature compression is avoided, in addition, a text feature processing component formed by a plurality of feature processing cores is adopted in the sliding window filtering (convolution processing/moving average processing) process, procedural information generated during sliding window filtering is not required to be reserved, and further, the occupation of storage resources of an AI cloud office server during cloud office scene session processing can be avoided as much as possible on the basis of ensuring cloud office scene session analysis quality.
For a description of the effects of the above AI cloud office server, computer-readable storage medium, see description of the above method.
In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are necessary for the embodiments to be used are briefly described below, the drawings being incorporated in and forming a part of the description, these drawings showing embodiments according to the present invention and together with the description serve to illustrate the technical solutions of the present invention. It is to be understood that the following drawings illustrate only certain embodiments of the invention and are therefore not to be considered limiting of its scope, for the person of ordinary skill in the art may admit to other equally relevant drawings without inventive effort.
Fig. 1 is a block diagram of an AI cloud office server according to an embodiment of the present invention.
Fig. 2 is a flow chart of a data processing method based on cloud office according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention.
Fig. 1 is a schematic structural diagram of an AI cloud office server 10 according to an embodiment of the present invention, including a processor 102, a memory 104, and a bus 106. The memory 104 is used for storing execution instructions, including a memory and an external memory, where the memory may also be understood as an internal memory, and is used for temporarily storing operation data in the processor 102 and data exchanged with the external memory such as a hard disk, where the processor 102 exchanges data with the external memory through the memory, and when the AI cloud office server 10 operates, the processor 102 and the memory 104 communicate with each other through the bus 106, so that the processor 102 executes the data processing method based on cloud office according to the embodiment of the present invention.
Referring to fig. 2, fig. 2 is a schematic flow chart of a cloud office based data processing method, which is applied to an AI cloud office server and may include steps 101-105.
And 101, obtaining an initial text semantic moment and a processed text semantic moment corresponding to the current cloud office scene session.
The current cloud office scene session and the initial text semantic moment (which can be understood as an initial session text feature diagram) both correspond to a first text granularity, the processed text semantic moment (which can be understood as a sampled session text feature diagram) corresponds to a second text granularity, and the second text granularity is smaller than the first text granularity.
In the embodiment of the invention, the AI cloud office server obtains a current cloud office scene session, and the current cloud office scene session can be a teleconference session or an office item request response dialogue, and the like, wherein the current cloud office scene session corresponds to a first text fine granularity. And carrying out sliding window filtering processing on the current cloud office scene session to obtain an initial text semantic moment under a first text granularity, and carrying out feature compression operation (which can be understood as downsampling processing) on the current cloud office scene session to obtain a processed text semantic moment under a second text granularity, wherein the second text granularity is smaller than the first text granularity.
In the embodiment of the invention, one AI expert system network according to the cloud office scene session disassembly project can comprise at least two network layers. One network layer is used for extracting text semantic moments under the first text granularity, and the other network layer is used for extracting text semantic moments under the second text granularity. And the AI cloud office server performs sliding window filtering on the current cloud office scene session to obtain an initial text semantic moment, and then performs feature compression on the initial text semantic moment to obtain a processed text semantic moment.
Step 102, obtaining a first text semantic moment through a first text feature processing component contained in an AI expert system network according to the initial text semantic moment.
Wherein the first text semantic moment corresponds to a first text granularity, the first text feature processing component comprises at least one feature processing core in cascade, each feature processing core for implementing a sliding window filtering process.
In the embodiment of the invention, the AI cloud office server loads the initial text semantic moment to a first text feature processing component (which can be understood as a residual component) in an AI expert system network, the first text semantic moment is output through the first text feature processing component, the text fine granularity of the first text semantic moment is the same as the first text fine granularity of the initial text semantic moment, and the first text feature processing component can be obtained by cascading at least one debugged feature processing core. The feature processing core may be a residual core, for example.
The AI cloud office server inputs the initial text semantic moment into a first text feature processing component to obtain a first text semantic moment, wherein the first text semantic moment corresponds to a first text fine granularity, namely is consistent with the initial text semantic moment and the text fine granularity of the current cloud office scene session, and the first text feature processing component comprises at least two feature processing cores in cascade connection.
And step 103, obtaining a second text semantic moment through a second text feature processing component contained in the AI expert system network according to the processed text semantic moment.
Wherein the second text semantic moment corresponds to a second text granularity, and the second text feature processing component comprises at least one feature processing core in cascade.
In the embodiment of the invention, the AI cloud office server loads the processing text semantic moment to the second text feature processing component in the AI expert system network, the second text semantic moment is output through the second text feature processing component, the text granularity of the second text semantic moment is the same as the second text granularity of the processing text semantic moment, and the second text feature processing component can be obtained by cascading at least one debugged feature processing core. The feature processing core may be a residual core, for example. The number of feature processing cores in the text feature processing component is not limited, and the more the number of feature processing cores is, the better the network performance is, and the system resource occupation of the AI cloud office server is not increased.
The AI cloud office server inputs the processing text semantic moment into a second text feature processing component to obtain a second text semantic moment, wherein the second text semantic moment corresponds to a second text fine granularity, namely is consistent with the text fine granularity of the processing text semantic moment, the second text feature processing component also comprises at least two cascaded feature processing cores, and in general, the network variable of the second text feature processing component is inconsistent with the network variable of the first text feature processing component.
Step 104, obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after the feature deriving operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after the feature compressing operation.
Wherein the third text semantic moment corresponds to the first text granularity and the fourth text semantic moment corresponds to the second text granularity.
In the embodiment of the invention, the AI cloud office server obtains a third text semantic moment by using the first text semantic moment and a second text semantic moment after feature derivation operation, and obtains a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after feature compression operation, wherein the text granularity of the third text semantic moment is the same as the text granularity of the initial text semantic moment, and the text granularity of the fourth text semantic moment is the same as the text granularity of the processed text semantic moment.
The AI cloud office server may combine the first text semantic moment and the second text semantic moment (which may also be understood as feature stacking processing) to obtain a third text semantic moment and a fourth text semantic moment respectively, where the third text semantic moment corresponds to the first text granularity and the fourth text semantic moment corresponds to the second text granularity.
And 105, obtaining a session analysis result corresponding to the current cloud office scene session through an AI expert system network according to the third text semantic moment and the fourth text semantic moment.
In the embodiment of the invention, the AI cloud office server performs combination processing on the third text semantic moment and the fourth text semantic moment to generate the target text semantic moment, loads the target text semantic moment into an AI expert system network, and outputs a corresponding session analysis result by using the item category through the AI expert system network. In the cloud office scene session tear down project, target session text including tear down from the current cloud office scene session may be output. And outputting the session risk theme corresponding to the current cloud office scene session in the cloud office scene session theme discrimination project.
Taking a cloud office scene session disassembly item as an example, after a third text semantic moment and a fourth text semantic moment are obtained, performing feature derivation operation on the fourth text semantic moment to obtain a text semantic moment consistent with the text fine granularity of the third text semantic moment, and then merging the text semantic moment with the third text semantic moment to obtain a target text semantic moment.
Taking a cloud office scene session topic discrimination item as an example, after a third text semantic moment and a fourth text semantic moment are obtained, performing feature compression operation on the third text semantic moment to obtain a text semantic moment consistent with the text granularity of the fourth text semantic moment, and then merging the text semantic moment with the fourth text semantic moment to obtain a target text semantic moment. And loading the target text semantic moment into a sliding window filtering unit (convolution unit), a downsampling unit (pooling unit) and a knowledge integration unit (full connection layer) in the AI expert system network, thereby obtaining the session risk discrimination viewpoint. For example, if a session risk discrimination view is made for a teleconference session, the session risk discrimination view may be a conference profile loss risk.
According to the data processing method based on cloud office, initial text semantic moment and processing text semantic moment corresponding to a current cloud office scene session can be obtained, then the first text semantic moment is obtained through a first text feature processing component contained in an AI expert system network according to the initial text semantic moment, then the second text semantic moment is obtained through a second text feature processing component contained in the AI expert system network according to the processing text semantic moment, then a third text semantic moment is obtained through the first text semantic moment and the second text semantic moment after feature deriving operation, a fourth text semantic moment is obtained through the second text semantic moment and the first text semantic moment after feature compressing operation, and finally a session analysis result corresponding to the current cloud office scene session is obtained through the AI expert system network according to the third text semantic moment and the fourth text semantic moment. Therefore, the text semantic moment with the initial high text granularity and the text semantic moment obtained by the feature compression can be subjected to detail mixing, the problem of detail loss caused by the feature compression is avoided, meanwhile, a text feature processing component formed by a plurality of feature processing cores is adopted in the sliding window filtering (convolution processing/reversible processing) process, procedural information generated during sliding window filtering is not required to be reserved, and further, on the basis of ensuring the cloud office scene session analysis quality, the occupation of storage resources of an AI cloud office server during cloud office scene session processing can be avoided as much as possible.
Under other possible design considerations, obtaining a third text semantic moment using the first text semantic moment and the second text semantic moment after the feature derivation operation, and obtaining a fourth text semantic moment using the second text semantic moment and the first text semantic moment after the feature compression operation may include: obtaining a first linked text semantic moment (a mixed text feature map) through a first sliding window filtering unit contained in an AI expert system network according to the first text semantic moment, wherein the first linked text semantic moment corresponds to a first text fine granularity; performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity; generating a third text semantic moment by using the first linkage text semantic moment and the second linkage text semantic moment; performing feature compression operation on the first text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity; according to the second text semantic moment, a fourth linkage text semantic moment is obtained through a fourth sliding window filtering unit contained in the AI expert system network, wherein the fourth linkage text semantic moment corresponds to the second text fine granularity; and generating a fourth text semantic moment by using the third linkage text semantic moment and the fourth linkage text semantic moment.
Or, obtaining the third text semantic moment by using the first text semantic moment and the second text semantic moment after the feature deriving operation, and obtaining the fourth text semantic moment by using the second text semantic moment and the first text semantic moment after the feature compressing operation may include: performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity; generating a third text semantic moment by using the first text semantic moment and the second linkage text semantic moment; performing feature compression operation on the second text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity; and generating a fourth text semantic moment by using the third linkage text semantic moment and the second text semantic moment.
In the concept of the text semantic moment aggregation processing, since the text granularity is different between the text semantic moments, the text granularity of the text semantic moments needs to be adjusted to ensure that the text granularity of the text semantic moments to be aggregated is the same. The aggregation processing of the text semantic moment can be realized by one of the following two ideas.
Thought 1: text semantic moments of the same text fine granularity are extracted through parallel sliding window filtering. And executing sliding window filtering processing on the first text semantic moment through a first sliding window filtering unit to obtain a first linkage text semantic moment. Because the text granularity of the second text semantic moment is smaller than that of the first text semantic moment, the second text semantic moment is required to be subjected to feature derivation through a second sliding window filtering unit to obtain a second linkage text semantic moment with the same text granularity as that of the first linkage text semantic moment, and therefore the first linkage text semantic moment and the second linkage text semantic moment can be combined to obtain a third text semantic moment. Correspondingly, the first text semantic moment is subjected to feature compression through a third sliding window filtering unit to obtain a third linkage text semantic moment. And carrying out sliding window filtering processing on the second text semantic moment through a fourth sliding window filtering unit to obtain a fourth linkage text semantic moment, wherein the third generation linkage text semantic moment is equal to the fourth linkage text semantic moment in text fine granularity. And combining the third linkage text semantic moment with the fourth linkage text semantic moment to obtain a fourth text semantic moment.
Idea 2: text semantic moments of the same text granularity are directly used. And carrying out feature derivation on the second text semantic moment through a second sliding window filtering unit to obtain a second linkage text semantic moment with the same text fine granularity as the first text semantic moment, so that the first text semantic moment and the second linkage text semantic moment can be combined to obtain a third text semantic moment. Correspondingly, feature compression is needed to be carried out on the third text semantic moment through a third sliding window filtering unit to obtain a third linkage text semantic moment with the same text fine granularity as that of the second text semantic moment, and therefore the second text semantic moment and the third linkage text semantic moment can be combined to obtain a fourth text semantic moment.
Therefore, the text semantic moment with high text granularity and the text semantic moment with low text granularity can be repeatedly aggregated, the features in the text semantic moment can repeatedly extract details from other features, the comprehensiveness and feature expressive performance of the text semantic moment are ensured, and the precision of a conversation analysis result is ensured.
In other possible embodiments, obtaining, by a first text feature processing component included in the AI expert system network, a first text semantic moment based on the initial text semantic moment may include: generating a first text semantic moment to be processed and a second text semantic moment to be processed by using the initial text semantic moment, wherein the sum of the description layer numbers of the first text semantic moment to be processed and the second text semantic moment to be processed is equal to the description layer (can be understood as a feature channel) number of the initial text semantic moment; carrying out multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through a first text feature processing component contained in an AI expert system network to obtain a first text convolution semantic moment; carrying out multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through a first text feature processing component contained in an AI expert system network to obtain a second text convolution semantic moment; and generating a first text semantic moment by using the first text convolution semantic moment and the second text convolution semantic moment, wherein the number of description layers of the first text semantic moment is equal to the sum of the number of description layers of the first text convolution semantic moment and the second text convolution semantic moment.
Illustratively, obtaining, by a second text feature processing component included in the AI expert system network, a second text semantic moment according to the processed text semantic moment may include: generating a third to-be-processed text semantic moment and a fourth to-be-processed text semantic moment by using the processing text semantic moment, wherein the sum of the description layer numbers of the third to-be-processed text semantic moment and the fourth to-be-processed text semantic moment is equal to the description layer number of the processing text semantic moment; performing multi-round sliding window filtering processing on the third text semantic moment to be processed through a second text feature processing component contained in the AI expert system network to obtain a third text convolution semantic moment; performing multi-round sliding window filtering processing on the fourth text semantic moment to be processed through a second text feature processing component contained in the AI expert system network to obtain a fourth text convolution semantic moment; and generating a second text semantic moment by using the third text convolution semantic moment and the fourth text convolution semantic moment, wherein the number of description layers of the second text semantic moment is equal to the sum of the number of description layers of the third text convolution semantic moment and the fourth text convolution semantic moment.
In the embodiment of the invention, an AI cloud office server firstly splits an initial text semantic moment into two sub-text semantic moments, namely a first text semantic moment to be processed and a second text semantic moment to be processed, and the sum of the number of the description layers of the two text semantic moments to be processed is equal to the number of the description layers of the initial text semantic moment. And respectively loading the first text semantic moment to be processed and the second text semantic moment to be processed into a first text feature processing component to carry out multi-round sliding window filtering, so as to obtain a first text convolution semantic moment and a second text convolution semantic moment. And finally integrating the two text convolution semantic moments to generate a first text semantic moment, wherein the number of the description layers of the first text semantic moment is equal to the number of the initial text semantic moments.
For the feature processing kernels obtained through cascading, the text convolution semantic moment C1 and the text convolution semantic moment C2 can be obtained through determining according to the text semantic moment M1 to be processed and the text semantic moment M2 to be processed. The input text semantic moment M1 to be processed and the text semantic moment M2 to be processed can be obtained through reverse determination according to the text convolution semantic moment C1 and the text convolution semantic moment C2. The intermediate data generated by the respective functions in the text feature processing component can be determined by the text semantic moment M1 to be processed and the text semantic moment M2 to be processed or by the text convolution semantic moment C1 and the text convolution semantic moment C2, whereby the intermediate data does not have to be retained by the text feature processing component. In this way, the text feature processing component obtained by cascading a plurality of feature processing cores can avoid retaining a large amount of intermediate data in the sliding window filtering process.
In an alternative embodiment, the performing, by using a first text feature processing component included in the AI expert system network, multiple sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed to obtain a first text convolution semantic moment may include: performing at least one round of sliding window filtering processing on the first text semantic moment to be processed through a first text feature processing component contained in an AI expert system network to obtain a first transition text semantic moment (which can be understood as an intermediate feature graph); the method comprises the steps that a whole downsampling unit contained in a first text feature processing component is used for downsampling a first transition text semantic moment to obtain first text semantic knowledge (which can be understood as a text knowledge vector or a text feature vector); processing the first text semantic knowledge through a knowledge integration unit contained in the first text feature processing component to obtain second text semantic knowledge, wherein the second text semantic knowledge is larger than the first text semantic knowledge in size; multiplying the first transition text semantic moment by adopting second text semantic knowledge to obtain a second transition text semantic moment; and obtaining a first text convolution semantic moment through a first text feature processing component contained in the AI expert system network according to the second transition text semantic moment and the second text semantic moment to be processed.
According to the method for processing the text semantic moment according to Attention Mechanism, the AI cloud office server firstly loads the first text semantic moment to be processed to the first text feature processing component to conduct least one-round sliding window filtering, so that a first transition text semantic moment is obtained, then loads the first transition text semantic moment to the integral downsampling unit to conduct downsampling processing, and first text semantic knowledge is obtained, namely dimension compression is conducted on the first transition text semantic moment. And the AI cloud office server loads the first text semantic knowledge to the knowledge integration unit for processing to obtain second text semantic knowledge, wherein the size of the second text semantic knowledge is larger than that of the first text semantic knowledge, namely the first text semantic knowledge is up-maintained. And carrying out dot product on the second text semantic knowledge and the first transition text semantic moment to obtain a second transition text semantic moment, and then loading the second transition text semantic moment and the second text semantic moment to be processed into the first text feature processing component to carry out sliding window filtering processing to obtain the first text convolution semantic moment.
In this way, a local attention mechanism is introduced into the feature processing core of the text feature processing component, and a weight is added to the description level of the text semantic moment, so that the features in the text semantic moment are more prominent, the matching of the local attention mechanism is realized, and the AI expert system network is enabled to master the text semantic moment with better quality.
In an alternative embodiment, according to the third text semantic moment and the fourth text semantic moment, obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session may include: acquiring linked text semantic moments of third text semantic moments through a fifth sliding window filtering unit contained in an AI expert system network, wherein the linked text semantic moments of the third text semantic moments correspond to the first text fine granularity; performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fourth text semantic moment, wherein the linked text semantic moment of the fourth text semantic moment corresponds to the first text fine granularity; generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment; generating a session disassembly result by using the target text semantic moment corresponding to the current cloud office scene session; and transmitting a session disassembly result to the session big data processing system so that the session big data processing system outputs the session disassembly result.
The embodiment of the invention provides a method for realizing cloud office scene session disassembly according to an AI expert system network. The AI cloud office server can aggregate the third text semantic moment and the fourth text semantic moment to obtain a session disassembly result. The text fine granularity of the fourth text semantic moment is smaller than that of the third text semantic moment, so that the AI cloud office server carries out sliding window filtering processing on the third text semantic moment through a fifth sliding window filtering unit to obtain a linked text semantic moment of the third text semantic moment, carries out feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit to obtain a linked text semantic moment of the sixth text semantic moment, and combines the two linked text semantic moments, thereby obtaining the target text semantic moment. And carrying out sliding window filtering processing according to the semantic moment of the target text, and generating a session disassembly result. The text granularity of the target text semantic moment is the same as the text granularity of the current cloud office scene session, and the text granularity of the target text semantic moment is the first text granularity. And finally, feeding back the session disassembly result to the session big data processing system, and outputting the session disassembly result by the session big data processing system.
Taking an AI expert system network (which can be a deep learning model) as an example, the text semantic moment T1 can obtain a corresponding linkage text semantic moment through a sliding window filtering unit, and the text semantic moment T1 can also be directly used as the linkage text semantic moment. And carrying out feature derivation operation on the text semantic moment T2 through a sliding window filtering unit to obtain the linkage text semantic moment corresponding to the text semantic moment T2. Similarly, the text semantic moment T3 performs feature derivation operation through a sliding window filtering unit to obtain linkage text semantic moment corresponding to the text semantic moment T3, and the three linkage text semantic moments are combined to obtain a target text semantic moment T4.
In the embodiment of the invention, a method for realizing cloud office scene session disassembly according to an AI expert system network is provided, wherein the AI expert system network fuses text semantic moments with low text fine granularity to text semantic moments with high text fine granularity, so that details contained in the obtained target text semantic moment are as complete as possible, and the accuracy and reliability of a disassembly result are ensured.
In an alternative embodiment, according to the third text semantic moment and the fourth text semantic moment, obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session may include: performing feature compression operation on the third text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the third text semantic moment, wherein the linked text semantic moment of the third text semantic moment corresponds to the second text fine granularity; acquiring linked text semantic moments of fourth text semantic moments through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the fourth text semantic moments correspond to the second text fine granularity; according to the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment, a first target text semantic moment is obtained through a ninth sliding window filtering unit contained in an AI expert system network; obtaining a second target text semantic moment through a downsampling unit contained in an AI expert system network according to the first target text semantic moment; according to the second target text semantic moment, acquiring a risk discrimination possibility map (risk topic probability distribution) corresponding to the current cloud office scene session through a knowledge integration unit contained in an AI expert system network; determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using the risk discrimination possibility map; and issuing a session risk judging viewpoint to the session big data processing system so that the session big data processing system outputs the session risk judging viewpoint.
The embodiment of the invention provides a method for judging session risk of a cloud office scene session according to an AI expert system network. The AI cloud office server may aggregate the third text semantic moment and the fourth text semantic moment to obtain a conversational risk discrimination perspective. The text fine granularity of the third text semantic moment is larger than that of the fourth text semantic moment, so that the AI cloud office server performs feature compression operation on the third text semantic moment through a seventh sliding window filtering unit to obtain a linked text semantic moment of the third text semantic moment, performs sliding window filtering processing on the fourth text semantic moment through an eighth sliding window filtering unit to obtain a linked text semantic moment of the fourth text semantic moment, combines the two linked text semantic moments, and performs sliding window filtering processing through a ninth sliding window filtering unit to obtain the first target text semantic moment. And inputting the first target text semantic moment into an average downsampling unit to obtain a second target text semantic moment. And finally, inputting the second target text semantic moment to a knowledge integration unit, and outputting a risk discrimination possibility map corresponding to the current cloud office scene session. And finally, feeding back the session risk discrimination viewpoint to the session big data processing system, and outputting the session risk discrimination viewpoint by the session big data processing system.
Taking an AI expert system network as an example, the text semantic moment T1 can perform feature compression operation through a sliding window filtering unit to obtain a linked text semantic moment of the text semantic moment T1, and the text semantic moment T2 can generate the linked text semantic moment through the sliding window filtering unit and can also be directly used as the linked text semantic moment. And aggregating the linked text semantic moment of the text semantic moment T1 and the linked text semantic moment of the text semantic moment T2 (or the text semantic moment T2) to obtain a target text semantic moment T5. The target text semantic moment T5 can be subjected to feature compression operation through the sliding window filtering unit to obtain the linkage text semantic moment of the target text semantic moment T5, and the text semantic moment T3 can be used for generating the linkage text semantic moment through the sliding window filtering unit and also can be directly used as the linkage text semantic moment. The method comprises the steps of aggregating the linkage text semantic moment of a target text semantic moment T5 and the linkage text semantic moment (or the text semantic moment T3) of the text semantic moment T3 to obtain a target text semantic moment T6, performing downsampling on the target text semantic moment T6 to obtain a target text semantic moment T7, and finally, processing the target text semantic moment T7 through a knowledge integration unit to obtain a risk discrimination possibility map.
In the embodiment of the invention, a method for judging the session risk of a cloud office scene session is provided according to an AI expert system network, wherein the AI expert system network compresses the text semantic moment characteristics with high text granularity after multidimensional aggregation and then adds the compressed text semantic moment characteristics with high text granularity into the text semantic moment with low text granularity, so that details contained in the text semantic moment with low text granularity are rich and complete as much as possible, and the accuracy and the credibility of the session risk judgment are ensured.
In an alternative embodiment, according to the third text semantic moment and the fourth text semantic moment, obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session may include: obtaining a fifth text semantic moment through a third text feature processing component contained in the AI expert system network according to the third text semantic moment, wherein the fifth text semantic moment corresponds to the first text fine granularity, and the third text feature processing component comprises at least one feature processing core in cascade; obtaining a sixth text semantic moment through a fourth text feature processing component contained in the AI expert system network according to the fourth text semantic moment, wherein the sixth text semantic moment corresponds to the second text fine granularity, and the fourth text feature processing component comprises at least one feature processing core in cascade; and according to the fifth text semantic moment and the sixth text semantic moment, obtaining a session analysis result corresponding to the current cloud office scene session through an AI expert system network.
After the third text semantic moment and the fourth text semantic moment are obtained, sliding window filtering processing can be continued. For example, after the third text semantic moment and the fourth text semantic moment are obtained in the manner of steps 101 to 104, the third text semantic moment is loaded to the third text feature processing component, the fifth text semantic moment is output by the third text feature processing component, and similarly, the fourth text semantic moment is loaded to the fourth text feature processing component, and the sixth text semantic moment is output by the fourth text feature processing component.
Taking a cloud office scene session disassembly item as an example, after a fifth text semantic moment and a sixth text semantic moment are obtained, performing feature derivation operation on the sixth text semantic moment to obtain a text semantic moment consistent with the text fine granularity of the fifth text semantic moment, and then merging the text semantic moment with the fifth text semantic moment to obtain a target text semantic moment.
Taking a cloud office scene session topic discrimination item as an example, after a fifth text semantic moment and a sixth text semantic moment are obtained, performing feature compression operation on the fifth text semantic moment to obtain a text semantic moment consistent with the text granularity of the sixth text semantic moment, and then merging the text semantic moment with the sixth text semantic moment to obtain a target text semantic moment. And loading the target text semantic moment to a sliding window filtering unit, a downsampling unit and a knowledge integration unit in the AI expert system network, so as to obtain a session risk discrimination viewpoint.
In the embodiment of the invention, a manner of generating text semantic moments with equal text granularity according to a plurality of text feature processing components is provided, so that sliding window filtering operation of the text semantic moments can be performed on the text semantic moments, on one hand, the depth of a network can be increased, more feature processing cores can be accumulated to obtain higher precision, and on the other hand, procedural information generated during sliding window filtering is not required to be reserved, so that resource expenditure required by cloud office scene conversation processing is reduced.
In an alternative embodiment, according to the fifth text semantic moment and the sixth text semantic moment, obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session may include: obtaining linked text semantic moments of a fifth text semantic moment through a fifth sliding window filtering unit contained in an AI expert system network, wherein the linked text semantic moments of the fifth text semantic moment correspond to the first text fine granularity; performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the sixth text semantic moment, wherein the linked text semantic moment of the sixth text semantic moment corresponds to the first text fine granularity; generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment; generating a session disassembly result by using the target text semantic moment corresponding to the current cloud office scene session; and transmitting a session disassembly result to the session big data processing system so that the session big data processing system outputs the session disassembly result.
In the embodiment of the invention, the AI cloud office server can aggregate the fifth text semantic moment and the sixth text semantic moment to obtain a session disassembly result. The text fine granularity of the sixth text semantic moment is smaller than that of the fifth text semantic moment, so that the AI cloud office server carries out sliding window filtering processing on the fifth text semantic moment through a fifth sliding window filtering unit to obtain a linkage text semantic moment of the fifth text semantic moment, carries out feature derivation operation on the sixth text semantic moment through the sixth sliding window filtering unit to obtain a linkage text semantic moment of the sixth text semantic moment, and combines the two obtained linkage text semantic moments to obtain the target text semantic moment. And carrying out sliding window filtering processing according to the semantic moment of the target text, and generating a session disassembly result. The text granularity of the target text semantic moment is the same as the text granularity of the current cloud office scene session, and the text granularity of the target text semantic moment is the first text granularity. And finally, feeding back the session disassembly result to the session big data processing system, and outputting the session disassembly result by the session big data processing system.
In the embodiment of the invention, another method for realizing cloud office scene session disassembly according to an AI expert system network is provided, so that sliding window filtering operation of a plurality of text feature processing components can be carried out on text semantic moments, on one hand, the depth of the network can be increased, more feature processing cores are accumulated to obtain higher precision, and the text semantic moments with low text fine granularity are fused to the text semantic moments with high text fine granularity, so that details contained in the obtained target text semantic moments are complete as much as possible, and the accuracy and reliability of a disassembly result are ensured.
In an alternative embodiment, according to the fifth text semantic moment and the sixth text semantic moment, obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session, including: performing feature compression operation on the fifth text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fifth text semantic moment, wherein the linked text semantic moment of the fifth text semantic moment corresponds to the second text fine granularity; obtaining a linked text semantic moment of a sixth text semantic moment through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moment of the sixth text semantic moment corresponds to the second text fine granularity; according to the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment, a first target text semantic moment is obtained through a ninth sliding window filtering unit contained in an AI expert system network; obtaining a second target text semantic moment through a downsampling unit contained in an AI expert system network according to the first target text semantic moment; according to the second target text semantic moment, acquiring a risk discrimination possibility map corresponding to the current cloud office scene session through a knowledge integration unit contained in an AI expert system network; determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using the risk discrimination possibility map; and issuing a session risk judging viewpoint to the session big data processing system so that the session big data processing system outputs the session risk judging viewpoint.
In the embodiment of the invention, another method for judging session risk of cloud office scene session according to an AI expert system network is provided. The AI cloud office server may fuse the fifth text semantic moment and the sixth text semantic moment to obtain a conversational risk discrimination perspective. The text fine granularity of the fifth text semantic moment is larger than that of the sixth text semantic moment, so that the AI cloud office server performs feature compression operation on the fifth text semantic moment through a seventh sliding window filtering unit to obtain a linkage text semantic moment of the fifth text semantic moment, performs sliding window filtering processing on the sixth text semantic moment through an eighth sliding window filtering unit to obtain a linkage text semantic moment of the sixth text semantic moment, combines the two obtained linkage text semantic moments, and performs sliding window filtering processing through a ninth sliding window filtering unit to obtain the first target text semantic moment. And inputting the first target text semantic moment into an average downsampling unit to obtain a second target text semantic moment. And finally, inputting the second target text semantic moment to a knowledge integration unit, and outputting a risk discrimination possibility map corresponding to the current cloud office scene session. And finally, feeding back the session risk discrimination viewpoint to the session big data processing system, and outputting the session risk discrimination viewpoint by the session big data processing system.
In the embodiment of the invention, another method for realizing session risk discrimination of cloud office scene session according to an AI expert system network is provided, so that sliding window filtering operation of a plurality of text feature processing components can be carried out on text semantic moments, on one hand, the depth of the network can be increased, more feature processing cores are accumulated to obtain higher precision, and the text semantic moment features with high text granularity after multidimensional aggregation are compressed and then fused to the text semantic moment with low text granularity, so that details contained in the text semantic moment with low text granularity are as rich and complete as possible, and the precision and reliability of session risk discrimination are ensured.
Based on the foregoing, a solution of network debugging in the present invention will be described, and one of the debugging steps of the AI expert system network includes steps 201-207.
Step 201, obtaining a sample cloud office scene session, where the sample cloud office scene session corresponds to an authenticated session disassembly result, and the authenticated session disassembly result is annotation information (training annotation) of the sample cloud office scene session on each text unit.
In the embodiment of the invention, the AI cloud office server obtains a sample cloud office scene session, which may be a teleconference session or an office item request response session, etc., where the sample cloud office scene session includes a real/correct disassembled result of the annotated cloud office scene session.
Step 202, obtaining a sample initial text semantic moment and a sample processed text semantic moment corresponding to a sample cloud office scene session through an AI expert system network, wherein the sample initial text semantic moment and the sample cloud office scene session both correspond to a first text granularity, the sample processed text semantic moment corresponds to a second text granularity, and the second text granularity is smaller than the first text granularity.
In the embodiment of the invention, an AI cloud office server performs sliding window filtering processing on a sample cloud office scene session through an AI expert system network to obtain a sample initial text semantic moment under a first text fine granularity, and performs feature compression operation on the sample cloud office scene session to obtain a sample processing text semantic moment under a second text fine granularity, wherein the second text fine granularity is smaller than the first text fine granularity.
Step 203, obtaining a first text semantic moment through a first text feature processing component included in the AI expert system network according to the sample initial text semantic moment, wherein the first text semantic moment corresponds to a first text fine granularity, and the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing.
In the embodiment of the invention, the AI cloud office server loads the initial text semantic moment of the sample to the first text feature processing component in the AI expert system network, the first text semantic moment is output through the first text feature processing component, the text fine granularity of the first text semantic moment of the sample is consistent with the first text fine granularity of the initial text semantic moment of the sample, and the first text feature processing component can be obtained by cascading the feature processing cores of at least one sample.
Step 204, according to the sample processing text semantic moment, obtaining a second sample text semantic moment through a second text feature processing component contained in the AI expert system network, wherein the second sample text semantic moment corresponds to a second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection.
In the embodiment of the invention, the AI cloud office server loads the sample processing text semantic moment to the second text feature processing component in the AI expert system network, the second sample text semantic moment is output through the second text feature processing component, the text granularity of the second sample text semantic moment is consistent with the second text granularity of the sample processing text semantic moment, and the second text feature processing component can be obtained by cascading the feature processing cores of at least one sample.
Step 205, obtaining a third sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment after the feature deriving operation, and obtaining a fourth sample text semantic moment by using the second sample text semantic moment and the first sample text semantic moment after the feature compressing operation, wherein the third sample text semantic moment corresponds to the first text granularity, and the fourth sample text semantic moment corresponds to the second text granularity.
In the embodiment of the invention, the AI cloud office server obtains a third sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment, and obtains a fourth sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment, wherein the text fine granularity of the third sample text semantic moment is the same as the text fine granularity of the sample initial text semantic moment, and the third sample text semantic moment and the second sample text semantic moment are both the first text fine granularity. The text granularity of the fourth sample text semantic moment is the same as the text granularity of the sample processing text semantic moment, and the fourth sample text semantic moment and the sample processing text semantic moment are both the second text granularity.
And 206, obtaining a session disassembly prediction result corresponding to the current cloud office scene session through an AI expert system network according to the third sample text semantic moment and the fourth sample text semantic moment.
In the embodiment of the invention, the AI cloud office server performs combination processing on the third sample text semantic moment and the fourth sample text semantic moment to generate the sample target text semantic moment, and loads the sample target text semantic moment to a processing unit and an output unit in an AI expert system network, so that a corresponding session disassembly prediction result is output, wherein the session disassembly prediction result can comprise target session text disassembled from a sample cloud office scene session.
Step 207, optimizing network variables of the AI expert system network by using the session disassembly prediction result and the authenticated session disassembly result through the network debugging cost, wherein the AI expert system network is the AI expert system network involved in the above embodiment.
In the embodiment of the invention, the AI cloud office server calculates the difference between the session disassembly prediction result and the authenticated session disassembly result by using the obtained session disassembly prediction result and the authenticated session disassembly result and uses the network debugging cost, so that the network variable of the AI expert system network is optimized by using the cost variable, and when the result of the network debugging cost tends to be stable, the AI expert system network can be considered to be debugged.
In the embodiment of the invention, a method for network debugging is provided, and an AI cloud office server utilizes a sample cloud office scene session corresponding to an authenticated session disassembly result to debug an AI expert system network. Therefore, the debugging data with different text fine granularity is utilized for debugging, and the accuracy of the cloud office scene session of the AI expert system network disassembly obtained by the debugging can be improved.
Another debugging step of the AI expert system network described above includes steps 301-307.
Step 301, obtaining a sample cloud office scene session, wherein the sample cloud office scene session corresponds to an authenticated session risk discrimination viewpoint, and the authenticated session risk discrimination viewpoint is information obtained by annotating a session risk topic of the sample cloud office scene session.
Step 302, obtaining a sample initial text semantic moment and a sample processed text semantic moment corresponding to a sample cloud office scene session through an AI expert system network, wherein the sample initial text semantic moment and the sample cloud office scene session both correspond to a first text granularity, the sample processed text semantic moment corresponds to a second text granularity, and the second text granularity is smaller than the first text granularity.
Step 303, obtaining a first text semantic moment through a first text feature processing component included in the AI expert system network according to the sample initial text semantic moment, wherein the first text semantic moment corresponds to a first text fine granularity, and the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing.
Step 304, according to the sample processing text semantic moment, obtaining a second sample text semantic moment through a second text feature processing component contained in the AI expert system network, wherein the second sample text semantic moment corresponds to a second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection.
Step 305, obtaining a third sample text semantic moment by using the first sample text semantic moment and the second sample text semantic moment after the feature deriving operation, and obtaining a fourth sample text semantic moment by using the second sample text semantic moment and the first sample text semantic moment after the feature compressing operation, wherein the third sample text semantic moment corresponds to the first text granularity, and the fourth sample text semantic moment corresponds to the second text granularity.
And step 306, obtaining a session risk prediction viewpoint corresponding to the current cloud office scene session through an AI expert system network according to the third text semantic moment and the fourth text semantic moment.
In the embodiment of the invention, the AI cloud office server combines the third sample text semantic moment and the fourth sample text semantic moment to generate the sample target text semantic moment, and loads the sample target text semantic moment to a sliding window filtering unit, a downsampling unit and a knowledge integration unit in an AI expert system network, thereby obtaining a session risk judging view. The session risk prediction perspective may include a session risk topic corresponding to the sample cloud office scene session.
Step 307, optimizing network variables of the AI expert system network by using the session risk prediction viewpoint and the authenticated session risk discrimination viewpoint through the network debugging cost, wherein the AI expert system network is the AI expert system network involved in the above embodiment.
In the embodiment of the invention, the AI cloud office server calculates the difference between the session risk prediction viewpoint and the authenticated session risk discrimination viewpoint by using the obtained cloud office scene session prediction discrimination viewpoint and the cloud office scene session real discrimination viewpoint and by using the network debugging cost, thereby optimizing the network variable of the AI expert system network by using the cost variable, and considering that the AI expert system network is debugged when the result of the network debugging cost tends to be stable.
In the embodiment of the invention, a method for network debugging is provided, and an AI cloud office server utilizes a sample cloud office scene session corresponding to a cloud office scene session real discrimination viewpoint to debug an AI expert system network. Therefore, the accuracy and the credibility of judging the session risk subject of the cloud office scene session by the AI expert system network obtained by debugging can be improved by utilizing the debugging data with different text fine granularity for debugging.
Further, there is also provided a readable storage medium having stored thereon a program which when executed by a processor implements the above-described method.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described system and apparatus may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein. In the several embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, and for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, device or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
Claims (9)
1. The data processing method based on cloud office is characterized by being applied to an AI cloud office server, and comprises the following steps:
obtaining an initial text semantic moment and a processed text semantic moment corresponding to a current cloud office scene session, wherein the current cloud office scene session and the initial text semantic moment both correspond to a first text fine granularity, the processed text semantic moment corresponds to a second text fine granularity, and the second text fine granularity is smaller than the first text fine granularity;
according to the initial text semantic moment, a first text semantic moment is obtained through a first text feature processing component contained in an AI expert system network, wherein the first text semantic moment corresponds to the first text fine granularity, the first text feature processing component comprises at least one feature processing core in cascade connection, and each feature processing core is used for realizing sliding window filtering processing;
obtaining a second text semantic moment through a second text feature processing component contained in the AI expert system network according to the processed text semantic moment, wherein the second text semantic moment corresponds to the second text fine granularity, and the second text feature processing component comprises at least one feature processing core in cascade connection;
Obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after feature derivation operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after feature compression operation, wherein the third text semantic moment corresponds to the first text granularity and the fourth text semantic moment corresponds to the second text granularity;
according to the third text semantic moment and the fourth text semantic moment, a session analysis result corresponding to the current cloud office scene session is obtained through the AI expert system network;
the method for obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after feature deriving operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after feature compressing operation comprises the following steps:
according to the first text semantic moment, a first linkage text semantic moment is obtained through a first sliding window filtering unit contained in the AI expert system network, wherein the first linkage text semantic moment corresponds to the first text fine granularity;
Performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity;
generating the third text semantic moment by using the first linkage text semantic moment and the second linkage text semantic moment;
performing feature compression operation on the first text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity;
obtaining a fourth linkage text semantic moment through a fourth sliding window filtering unit contained in the AI expert system network according to the second text semantic moment, wherein the fourth linkage text semantic moment corresponds to the second text fine granularity;
generating a fourth text semantic moment by using the third linkage text semantic moment and the fourth linkage text semantic moment;
or, the obtaining a third text semantic moment by using the first text semantic moment and the second text semantic moment after the feature deriving operation, and obtaining a fourth text semantic moment by using the second text semantic moment and the first text semantic moment after the feature compressing operation, including:
Performing feature derivation operation on the second text semantic moment through a second sliding window filtering unit contained in the AI expert system network to obtain a second linkage text semantic moment, wherein the second linkage text semantic moment corresponds to the first text fine granularity;
generating the third text semantic moment by using the first text semantic moment and the second linkage text semantic moment;
performing feature compression operation on the first text semantic moment through a third sliding window filtering unit contained in the AI expert system network to obtain a third linkage text semantic moment, wherein the third linkage text semantic moment corresponds to the second text fine granularity;
and generating the fourth text semantic moment by using the third linkage text semantic moment and the second text semantic moment.
2. The method of claim 1, wherein said obtaining a first text semantic moment via a first text feature processing component comprised by an AI expert system network based on said initial text semantic moment comprises:
generating a first text semantic moment to be processed and a second text semantic moment to be processed by using the initial text semantic moment, wherein the sum of the description layer numbers of the first text semantic moment to be processed and the second text semantic moment to be processed is equal to the description layer number of the initial text semantic moment;
Performing multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through the first text feature processing component contained in the AI expert system network to obtain a first text convolution semantic moment;
performing multi-round sliding window filtering processing on the first text semantic moment to be processed and the second text semantic moment to be processed through the first text feature processing component contained in the AI expert system network to obtain a second text convolution semantic moment;
generating the first text semantic moment by using the first text convolution semantic moment and the second text convolution semantic moment, wherein the number of description layers of the first text semantic moment is equal to the sum of the numbers of description layers of the first text convolution semantic moment and the second text convolution semantic moment;
the step of obtaining a second text semantic moment through a second text feature processing component included in the AI expert system network according to the processed text semantic moment includes:
generating a third to-be-processed text semantic moment and a fourth to-be-processed text semantic moment by using the processing text semantic moment, wherein the sum of the description layer numbers of the third to-be-processed text semantic moment and the fourth to-be-processed text semantic moment is equal to the description layer number of the processing text semantic moment;
Performing multi-round sliding window filtering processing on the third text semantic moment to be processed through the second text feature processing component contained in the AI expert system network to obtain a third text convolution semantic moment;
performing multi-round sliding window filtering processing on the fourth text semantic moment to be processed through the second text feature processing component contained in the AI expert system network to obtain a fourth text convolution semantic moment;
and generating the second text semantic moment by using the third text convolution semantic moment and the fourth text convolution semantic moment, wherein the number of description layers of the second text semantic moment is equal to the sum of the number of description layers of the third text convolution semantic moment and the fourth text convolution semantic moment.
3. The method of claim 2, wherein the performing, by the first text feature processing component included in the AI expert system network, a multi-pass sliding window filtering process on the first text semantic moment to be processed and the second text semantic moment to be processed to obtain a first text convolution semantic moment, includes:
the first text feature processing component contained in the AI expert system network is used for carrying out least one-round sliding window filtering processing on the first text semantic moment to be processed to obtain a first transition text semantic moment;
The first transition text semantic moment is subjected to downsampling processing through an integral downsampling unit contained in the first text feature processing assembly, so that first text semantic knowledge is obtained;
processing the first text semantic knowledge through a knowledge integration unit contained in the first text feature processing component to obtain second text semantic knowledge, wherein the size of the second text semantic knowledge is larger than that of the first text semantic knowledge;
multiplying the first transition text semantic moment by adopting the second text semantic knowledge to obtain a second transition text semantic moment;
and obtaining the first text convolution semantic moment through the first text feature processing component contained in the AI expert system network according to the second transition text semantic moment and the second text semantic moment to be processed.
4. The method of claim 1, wherein the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment comprises:
acquiring linked text semantic moments of the third text semantic moment through a fifth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the third text semantic moment correspond to the first text fine granularity;
Performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fourth text semantic moment, wherein the linked text semantic moment of the fourth text semantic moment corresponds to the first text fine granularity;
generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment;
generating a session disassembly result by utilizing the target text semantic moment corresponding to the current cloud office scene session;
and transmitting the session disassembly result to a session big data processing system so that the session big data processing system outputs the session disassembly result.
5. The method of claim 1, wherein the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment comprises:
performing feature compression operation on the third text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the third text semantic moment, wherein the linked text semantic moment of the third text semantic moment corresponds to the second text fine granularity;
Acquiring linked text semantic moments of the fourth text semantic moment through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the fourth text semantic moment correspond to the second text fine granularity;
acquiring a first target text semantic moment through a ninth sliding window filtering unit contained in the AI expert system network according to the linked text semantic moment of the third text semantic moment and the linked text semantic moment of the fourth text semantic moment;
obtaining a second target text semantic moment through a downsampling unit contained in the AI expert system network according to the first target text semantic moment;
according to the second target text semantic moment, acquiring a risk discrimination possibility map corresponding to the current cloud office scene session through a knowledge integration unit contained in the AI expert system network;
determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using a risk discrimination possibility map;
and issuing the session risk discrimination viewpoint to a session big data processing system so that the session big data processing system outputs the session risk discrimination viewpoint.
6. The method of claim 1, wherein the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the third text semantic moment and the fourth text semantic moment comprises:
obtaining a fifth text semantic moment through a third text feature processing component contained in the AI expert system network according to the third text semantic moment, wherein the fifth text semantic moment corresponds to the first text fine granularity, and the third text feature processing component comprises at least one feature processing core in cascade connection;
obtaining a sixth text semantic moment through a fourth text feature processing component contained in the AI expert system network according to the fourth text semantic moment, wherein the sixth text semantic moment corresponds to the second text fine granularity, and the fourth text feature processing component comprises at least one feature processing core in cascade connection;
and obtaining a session analysis result corresponding to the current cloud office scene session through the AI expert system network according to the fifth text semantic moment and the sixth text semantic moment.
7. The method of claim 6, wherein the obtaining, by the AI expert system network, a session analysis result corresponding to the current cloud office scene session according to the fifth text semantic moment and the sixth text semantic moment comprises:
Acquiring linked text semantic moments of the fifth text semantic moment through a fifth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the fifth text semantic moment correspond to the first text fine granularity; performing feature derivation operation on the fourth text semantic moment through a sixth sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the sixth text semantic moment, wherein the linked text semantic moment of the sixth text semantic moment corresponds to the first text fine granularity; generating a target text semantic moment corresponding to the current cloud office scene session by utilizing the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment; generating a session disassembly result by utilizing the target text semantic moment corresponding to the current cloud office scene session; the session risk judging view is issued to a session big data processing system, so that the session big data processing system outputs the session dismantling result;
or alternatively, the process may be performed,
performing feature compression operation on the fifth text semantic moment through a seventh sliding window filtering unit contained in the AI expert system network to obtain a linked text semantic moment of the fifth text semantic moment, wherein the linked text semantic moment of the fifth text semantic moment corresponds to the second text fine granularity; acquiring linked text semantic moments of the sixth text semantic moment through an eighth sliding window filtering unit contained in the AI expert system network, wherein the linked text semantic moments of the sixth text semantic moment correspond to the second text fine granularity; obtaining a first target text semantic moment through a ninth sliding window filtering unit contained in the AI expert system network according to the linked text semantic moment of the fifth text semantic moment and the linked text semantic moment of the sixth text semantic moment; obtaining a second target text semantic moment through a downsampling unit contained in the AI expert system network according to the first target text semantic moment; according to the second target text semantic moment, acquiring a risk discrimination possibility map corresponding to the current cloud office scene session through a knowledge integration unit contained in the AI expert system network; determining a session risk discrimination viewpoint corresponding to the current cloud office scene session by using a risk discrimination possibility map; and issuing the session risk discrimination viewpoint to a session big data processing system so that the session big data processing system outputs the session risk discrimination viewpoint.
8. An AI cloud office server is characterized by comprising a processor and a memory; the processor being communicatively connected to the memory, the processor being adapted to read a computer program from the memory and execute it to carry out the method of any of the preceding claims 1-7.
9. A computer readable storage medium, characterized in that it has stored thereon a computer program, which, when run, implements the method of any of the preceding claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310903597.XA CN116611799B (en) | 2023-07-22 | 2023-07-22 | Cloud office-based data processing method, AI cloud office server and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310903597.XA CN116611799B (en) | 2023-07-22 | 2023-07-22 | Cloud office-based data processing method, AI cloud office server and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116611799A CN116611799A (en) | 2023-08-18 |
| CN116611799B true CN116611799B (en) | 2023-09-26 |
Family
ID=87682329
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310903597.XA Active CN116611799B (en) | 2023-07-22 | 2023-07-22 | Cloud office-based data processing method, AI cloud office server and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116611799B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409016A (en) * | 2021-06-25 | 2021-09-17 | 东莞市镁客教育科技有限公司 | Information processing method, server and medium applied to big data cloud office |
| CN114186607A (en) * | 2021-11-03 | 2022-03-15 | 张俊杰 | Big data processing method and artificial intelligence server applied to cloud office |
| CN114186273A (en) * | 2021-12-10 | 2022-03-15 | 天津痴凡互联网科技有限公司 | Information security analysis method based on big data office and storage medium |
| CN114661572A (en) * | 2022-03-31 | 2022-06-24 | 杨喜顺 | Big data analysis method and system for user habit in handling digital cloud office |
| CN116362261A (en) * | 2023-04-10 | 2023-06-30 | 昆明谐善科技有限公司 | User session information analysis method and software product for dealing with digital service items |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230092702A1 (en) * | 2021-09-20 | 2023-03-23 | Salesforce, Inc. | Database systems and methods of naming record groups |
-
2023
- 2023-07-22 CN CN202310903597.XA patent/CN116611799B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409016A (en) * | 2021-06-25 | 2021-09-17 | 东莞市镁客教育科技有限公司 | Information processing method, server and medium applied to big data cloud office |
| CN114186607A (en) * | 2021-11-03 | 2022-03-15 | 张俊杰 | Big data processing method and artificial intelligence server applied to cloud office |
| CN114186273A (en) * | 2021-12-10 | 2022-03-15 | 天津痴凡互联网科技有限公司 | Information security analysis method based on big data office and storage medium |
| CN114661572A (en) * | 2022-03-31 | 2022-06-24 | 杨喜顺 | Big data analysis method and system for user habit in handling digital cloud office |
| CN116362261A (en) * | 2023-04-10 | 2023-06-30 | 昆明谐善科技有限公司 | User session information analysis method and software product for dealing with digital service items |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116611799A (en) | 2023-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111858869B (en) | Data matching method and device, electronic equipment and storage medium | |
| CN110570346B (en) | Method for performing style migration on calligraphy based on cyclic generation countermeasure network | |
| CN112632046A (en) | Cloud rule engine implementation method, system, device and medium | |
| CN102984050A (en) | Method, client and system for searching voices in instant messaging | |
| CN110852295B (en) | Video behavior recognition method based on multitasking supervised learning | |
| Dai et al. | A choreography analysis approach for microservice composition in cyber-physical-social systems | |
| CN111414842B (en) | Video comparison method and device, computer equipment and storage medium | |
| CN107784327A (en) | A kind of personalized community discovery method based on GN | |
| CN112581578A (en) | Cloud rendering system based on software definition | |
| CN113963170A (en) | RGBD image saliency detection method based on interactive feature fusion | |
| CN111860853A (en) | Online prediction system, online prediction equipment, online prediction method and electronic equipment | |
| CN116611799B (en) | Cloud office-based data processing method, AI cloud office server and medium | |
| CN112818958A (en) | Action recognition method, device and storage medium | |
| CN113010469A (en) | Image feature extraction method, device and computer-readable storage medium | |
| CN112434798A (en) | Multi-scale image translation method based on semi-supervised learning | |
| CN110491386A (en) | A kind of method, apparatus and computer readable storage medium generating meeting summary | |
| CN115858322A (en) | Log data processing method and device and computer equipment | |
| CN112990370B (en) | Image data processing method and device, storage medium and electronic equipment | |
| CN112529064B (en) | Efficient real-time semantic segmentation method | |
| CN114579634A (en) | Information mining method and system combining artificial intelligence and big data | |
| CN114356512A (en) | Data processing method, data processing equipment and computer readable storage medium | |
| CN113805966A (en) | Multi-terminal device dynamic view rendering implementation method and system | |
| CN106776947A (en) | Resource acquiring method, device and terminal | |
| CN112988724A (en) | Work efficiency data acquisition method and system | |
| Zheng et al. | CAILA: Concept-Aware Intra-Layer Adapters for Compositional Zero-Shot Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |