CN110059189A - A kind of categorizing system and method for gaming platform message - Google Patents
A kind of categorizing system and method for gaming platform message Download PDFInfo
- Publication number
- CN110059189A CN110059189A CN201910289668.5A CN201910289668A CN110059189A CN 110059189 A CN110059189 A CN 110059189A CN 201910289668 A CN201910289668 A CN 201910289668A CN 110059189 A CN110059189 A CN 110059189A
- Authority
- CN
- China
- Prior art keywords
- message
- rubbish
- model
- user
- sample
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The invention discloses a kind of message category system and method for gaming platform, system includes rubbish message discovery module, sample database, model training module and real-time processing module.Rubbish message discovery module obtains sample message by PU learning algorithm process rubbish message and is stored in sample database；Model training module uses the sample message of sample database, and according to the online period training similarity disaggregated model of game or character level CNN textual classification model；The disaggregated model of real-time processing module calling model training module differentiates user message, and carries out secondary discrimination by LP algorithm and key word analysis, and result is fed back to rubbish message categorization module.To which PU learning algorithm realizes that a small amount of sample can train, and solve the problems, such as that Cold Start lacks sample；By character level CNN textual classification model, recall rate and compatibility are improved；By LP algorithm and key word analysis, Real-time defence is realized, while obtaining new sample message building similarity model, solve the problems, such as cold start-up.
Description
Technical field
The present invention relates to data processing fields, particularly relate to a kind of gaming platform message category system and method.
Background technique
With the gradually development of game industry, the case where chat interface is flooded with rubbish message inside online game, is also more next
It is more, how to be classified (rubbish message is fallen in automatic fitration) to the message of gaming platform, prevent user by malice advertisement brush screen,
It is the problem of each game company will face that abuse even generation, which fills fraud information deception,.Traditional solution is according to different trips
The difference of company technique strength of playing and scale of construction size, generally has following several:
1, keyword shields: setting the keyword of corresponding rubbish message, identification classification is carried out to the message of user, after triggering
Rubbish message is closed；
2, manual title: operation personnel searches rubbish message and carries out title operation；
3, it increases speech difficulty: such as adjusting the grade that can make a speech, stage property speech etc.；
4, supervision type machine learning model is constructed: collection sample training message category model (such as the prison such as Bayes, SVM
Superintend and direct learning model).
Most of game company can make the chat about games message category system of itself, Preliminary Exploitation fortune in conjunction with aforesaid way
Battalion personnel close and rubbish message is collected and (optionally increases speech difficulty), then extract keyword and fill black name
It is singly shielded, has the company of certain technical conditions that supervised learning algorithm can be selected to construct message category model, to reach rubbish
The classification shielding of rubbish message to a certain extent.
However keyword shielding can only shield existing keyword (administrative staff collect manually), but malice is played
The expression way of rubbish breath can be changed after family's discovery targeted specifically, game company can only passively add blacklist, be difficult into
Row Real-time defence.Machine learning model has the expression way of certain learning ability prevention malicious players change rubbish message, but
Do so the problem of can not being fully solved Real-time defence because the effect of supervision type machine learning model depend on sample set with
And subsequent feature extraction, thus need to carry out sample collection and Feature Engineering, how to be labeled to data set is to need
The problem of facing (data set mark collects extremely labor intensive).If market is just invested in a game, in no any sample
In the case where this, emerging rubbish message can be described as helpless, sample is not enough to train new model first, black
List mode real-time is again too poor, is also to be badly in need of solving for this kind of cold start-up problem.In addition, existing categorizing system is expansible
Property not strong, such as some big game companies, under possess several money game or multilingual version has been issued in a game, it is right
In the multilingual situation of this more game, it may be necessary to the adaptable categorizing system of the more sets of building.
To sum up, current gaming platform message category method has following bigger defect:
1. sample collection problem: supervision type learning algorithm relies on great amount of samples；
Think over a problem 2. classifying quality is paid no attention to: existing model scheme effect is less desirable；
3. compatibility issue: needing a set of classification schemes for being suitable for any game and language；
4. new type rubbish message Real-time defence problem: needing to identify emerging rubbish message；
5. game is cold-started problem: how the game incipient stage carries out gaming platform message category in time.
In view of this, the present inventor does not attain regarding to the issue above improves caused many missings and inconvenience, and go deep into structure
Think, and actively research improvement has a fling at and develops and design the present invention.
Summary of the invention
The purpose of the present invention is to provide a kind of message category system and methods of gaming platform, solve existing message category
System there are the drawbacks of, have the characteristics that a small amount of sample can train, recall rate is high, compatibility is strong, realize Real-time defence, solve
Cold start-up problem.
In order to achieve the above objectives, solution of the invention is:
A kind of message category system of gaming platform, including rubbish message discovery module, sample database, model training module and
Real-time processing module；The rubbish message discovery module passes through customer service platform, data warehouse, user's report and real-time processing module
Obtain the rubbish message of gaming platform；Rubbish message categorization module by PU learning algorithm to the rubbish message of acquisition into
Row processing, obtains sample message and is stored in sample database；The model training module uses the sample message of sample database, and according to trip
Online period of playing is trained disaggregated model, and disaggregated model includes similarity disaggregated model and character level CNN text classification mould
Type；Game initial go-live period, model training module training similarity disaggregated model；Game online middle and later periods, model training module instruction
Practice character level CNN textual classification model；The disaggregated model of the real-time processing module calling model training module is to user message
It carries out differentiating normal or rubbish message, and the user message for being determined as rubbish message is fed back into rubbish message discovery module；It is real
When processing module be equipped with blacklist, the special word of blacklist is configured by the administrative staff of gaming platform；The real-time processing
Module carries out secondary discrimination to the user message for being determined as normal messages by LP algorithm and key word analysis, passes through key word analysis
The user message of keyword exception is labeled as rubbish message, passes through the type of the user message of LP algorithm tag keyword exception.
The rubbish message that the customer service platform storage contact staff collects；The institute stored in data warehouse storage gaming platform
There is user message；User provides rubbish message by way of user's report in game process for message category system.
The similarity disaggregated model is classified to calculate the distance between training sample message.
The character level CNN textual classification model is made of 6 convolutional layers and 3 full articulamentums；The input layer connects
User message is received, and inputs character level CNN text classification after carrying out one-hot coding to user message according to the size of alphabet
Model, CNN textual classification model carry out that word insertion conversion forming shape is (None, 1014,128) to the character after coding
Amount, then enter data into convolutional layer；Each convolutional layer carries out one-dimensional convolution to data, then carries out pond, finally obtains shape
For the tensor of (None, 34,256), then enter data into full articulamentum；Full articulamentum convert the data into shape be (None,
8704) tensor, and dropout layers of progress model regularization are set between two neighboring full articulamentum；Output layer passes through
The probability of classification described in softmax output input character, provides classification results.
A kind of message category method of gaming platform, using a kind of message category system of gaming platform, including
Following steps:
Step 1, system receive real-time user message, disaggregated model are judged whether there is, if so, going to step 2；If
It is no, then go to step 3；
The disaggregated model of step 2, real-time processing module calling model training module, classifies to user message, if
Rubbish message, then export rubbish message as a result, this user message processing terminate；If normal messages then go to step 3；
Step 3, real-time processing module differentiate user message by LP algorithm and key word analysis, judge keyword
It is whether abnormal, if it is not, then export normal messages as a result, this user message processing terminate；If so, passing through key word analysis
User message is labeled as rubbish message, feeds back to rubbish message by the type of LP algorithm tag user message, and by result
Discovery module；
Step 4, rubbish message discovery module are handled user message and its type by PU learning algorithm,
It obtains new sample message and updates sample database；Model training module is trained using the sample message of sample database and is updated point
Class model.
After above system and method, the present invention utilizes PU learning algorithm, only needs a small amount of sample that can be instructed
Practice, while solving the problems, such as that Cold Start lacks sample；Based on word component classification model, pass through character level CNN text point
Class model greatly improves recall rate, and can be compatible with more game or multilingual gaming platform；Pass through LP algorithm and pass
Keyword analysis, accomplishes Real-time defence to the unrecognized rubbish message of model and emerging rubbish message, and prevent in real time
Sample message is obtained in imperial process, is then constructed similarity model and is on the defensive, to solve the problems, such as cold start-up.
Detailed description of the invention
Fig. 1 is system structure of the invention schematic diagram；
Fig. 2 is the structural schematic diagram of character level CNN textual classification model；
Fig. 3 is the process schematic of present invention processing user message.
Specific embodiment
In order to further explain the technical solution of the present invention, below by specific embodiment, the present invention will be described in detail.
As shown in Figure 1, the present invention is a kind of message category system of gaming platform, including rubbish message discovery module, sample
This library, model training module and real-time processing module.
Above-mentioned rubbish message discovery module passes through the way such as customer service platform, data warehouse, user's report and real-time processing module
The rubbish message of diameter acquisition gaming platform；Rubbish message categorization module is by PU learning algorithm to the rubbish message of acquisition
It is handled, obtain sample message and is stored in sample database.The rubbish message that above-mentioned customer service platform storage contact staff collects；Data
All user messages stored in warehouse storage gaming platform, can therefrom extract rubbish message；User is led in game process
It crosses the form of user's report and provides rubbish message for message category system.
Above-mentioned model training module uses the sample message of sample database, and is trained classification according to game online period
Model, disaggregated model include similarity disaggregated model and character level CNN textual classification model.Game initial go-live period, model training
Module trains similarity disaggregated model, and similarity disaggregated model is classified to calculate the distance between training sample message；
Game online middle and later periods, model training module training character level CNN textual classification model.
Game initial go-live period: rubbish message is few in game, and it is each to can use similarity calculation for only a small amount of sample message
The distance between a sample message is classified.For any one sample message, its benefit can be expressed as a vectorFor any one user message, can be expressed asThat
According to cosine similarity, the distance between available any one user message and any one sample message are as follows:According to apart from size whether be greater than threshold value can be carried out judging user input into
Whether row is rubbish message, and it is temporary can to generate a simple and reliable rubbish message disaggregated model in a short time for we in this way
When cope with game initial waste message problem.
The game online later period: as rubbish message is more and more, the quantity of sample message increases, and finds in conjunction with rubbish message
Module utilizes training sample set training character level CNN textual classification model, the structural representation of character level CNN textual classification model
Figure as shown in Fig. 2, character level CNN textual classification model includes input layer, 6 convolutional layers, 3 full articulamentums and output layer, this
When the input space be exactly all user messages set, output space be exactly rubbish message and normal messages set, each
User message is exactly a message instance.Input layer receives user message, and is carried out according to the size of alphabet to user message
Character level CNN textual classification model is inputted after one-hot coding, it is embedding that CNN textual classification model carries out word to the character after coding
Enter to convert forming shape and be the tensor of (None, 1014,128), then enters data into convolutional layer；Each convolutional layer carries out one to data
Convolution is tieed up, pond is then carried out, shape is finally obtained and is the tensor of (None, 34,256), then enter data into full articulamentum；Entirely
Articulamentum converts the data into shape and is the tensor of (None, 8704), and dropout is arranged between two neighboring full articulamentum
Layer carries out model regularization；Output layer exports the probability of classification described in input character by softmax, provides classification results.Mould
The each step details of type is as follows:
Such as the step of being classified using character level CNN textual classification model to message " filling 100 send 500vip to supplement with money "
It is as follows:
1. pair message carries out one-hot coding, it is converted into following vector:
2. being exported by convolutional layer and full articulamentum are as follows: [[0.659829850.34017012]], wherein
0.65982985 expression is the probability of rubbish message, and 0.34017012 expression is the probability of normal messages.
3. exporting the message is advertisement prompting, such as: " message: filling 100 500vip is sent to supplement with money advertisement is judged as by model ".
Using character level CNN textual classification model, text feature can be automatically extracted by neural network in the training process
Out, the compatibility for allowing for system in this way is extremely strong, no matter which kind of game or which kind of language can have good effect, and
And recall rate has greatly improved compared to tradition and its learning classification model.
The disaggregated model of above-mentioned real-time processing module calling model training module sentences the user message of gaming platform
Other normal messages or rubbish message, and the user message for being determined as rubbish message is fed back into rubbish message discovery module, with this
Realize the rubbish message kept out in most of live chat in real time.Real-time processing module is equipped with blacklist, the special word of blacklist
Remittance is configured by the administrative staff of gaming platform, to prevent special word that may be present around disaggregated model.It is above-mentioned real-time
Processing module carries out secondary discrimination to the user message for being determined as normal messages by LP algorithm and key word analysis, passes through key
The user message of keyword exception is labeled as rubbish message by word analysis, passes through the user message of LP algorithm tag keyword exception
Type.
When a novel rubbish message bypass disaggregated model and blacklist (being mistaken for normal messages), in real time at this time
It is rubbish message that processing module can mark automatically this message by key word analysis, then passes through LP algorithm for this message
Type, which auto marks out, to be come, to be on the defensive to this new type rubbish message.
As shown in figure 3, being based on above system, a kind of message category method of gaming platform, comprising the following steps:
Step 1, system receive real-time user message, disaggregated model are judged whether there is, if so, going to step 2；If
It is no, then go to step 3；
The disaggregated model of step 2, real-time processing module calling model training module, classifies to user message, if
Rubbish message, then export rubbish message as a result, this user message processing terminate；If normal messages then go to step 3；
Step 3, real-time processing module differentiate user message by LP algorithm and key word analysis, judge keyword
It is whether abnormal, if it is not, then export normal messages as a result, this user message processing terminate；If so, passing through key word analysis
User message is labeled as rubbish message, feeds back to rubbish message by the type of LP algorithm tag user message, and by result
Discovery module；
Step 4, rubbish message discovery module are handled user message and its type by PU learning algorithm,
It obtains new sample message and updates sample database；Model training module is trained using the sample message of sample database and is updated point
Class model.
By the systems and methods, the present invention utilizes PU learning algorithm, only needs a small amount of sample that can be instructed
Practice, while solving the problems, such as that Cold Start lacks sample；Based on word component classification model, pass through character level CNN text point
Class model greatly improves recall rate, and can be compatible with more game or multilingual gaming platform；Pass through LP algorithm and pass
Keyword analysis, accomplishes Real-time defence to the unrecognized rubbish message of model and emerging rubbish message, and prevent in real time
Sample message is obtained in imperial process, is then constructed similarity model and is on the defensive, to solve the problems, such as cold start-up.
Above-described embodiment and schema and non-limiting product form and style of the invention, any technical field it is common
The appropriate changes or modifications that technical staff does it all should be regarded as not departing from patent category of the invention.
Claims (5)
1. a kind of message category system of gaming platform, it is characterised in that: including rubbish message discovery module, sample database, model
Training module and real-time processing module；The rubbish message discovery module passes through customer service platform, data warehouse, user's report and reality
When processing module obtain gaming platform rubbish message；Rubbish message categorization module is by PU learning algorithm to acquisition
Rubbish message is handled, and is obtained sample message and is stored in sample database；
The model training module uses the sample message of sample database, and is trained classification mould according to game online period
Type, disaggregated model include similarity disaggregated model and character level CNN textual classification model；Game initial go-live period, model training mould
Block trains similarity disaggregated model；Game online middle and later periods, model training module training character level CNN textual classification model；
The disaggregated model of the real-time processing module calling model training module carries out user message to differentiate that normal or rubbish disappears
Breath, and the user message for being determined as rubbish message is fed back into rubbish message discovery module；Real-time processing module is equipped with blacklist,
The special word of blacklist is configured by the administrative staff of gaming platform；The real-time processing module passes through LP algorithm and key
Word analysis carries out secondary discrimination to the user message for being determined as normal messages, by key word analysis by the user of keyword exception
Message is labeled as rubbish message, passes through the type of the user message of LP algorithm tag keyword exception.
2. a kind of message category system of gaming platform as described in claim 1, it is characterised in that: the customer service platform storage
The rubbish message that contact staff collects；All user messages stored in data warehouse storage gaming platform；User is in game
Rubbish message is provided for message category system by way of user's report in journey.
3. a kind of message category system of gaming platform as described in claim 1, it is characterised in that: the similarity classification mould
Type is classified to calculate the distance between training sample message.
4. a kind of message category system of gaming platform as described in claim 1, it is characterised in that: the character level CNN text
This disaggregated model is made of 6 convolutional layers and 3 full articulamentums；The input layer receives user message, and according to alphabet
Size to user message carry out one-hot coding after input character level CNN textual classification model, CNN textual classification model pair
Character after coding carries out word insertion conversion forming shape and is the tensor of (None, 1014,128), then enters data into convolutional layer；Often
A convolutional layer carries out one-dimensional convolution to data, then carries out pond, finally obtains the tensor that shape is (None, 34,256), then
Enter data into full articulamentum；Full articulamentum converts the data into the tensor that shape is (None, 8704), and two neighboring complete
Dropout layers of progress model regularization are set between articulamentum；Output layer exports classification described in input character by softmax
Probability provides classification results.
5. a kind of message category method of gaming platform uses a kind of message of any gaming platform of Claims 1-4 4
Categorizing system, it is characterised in that the following steps are included:
Step 1, system receive real-time user message, disaggregated model are judged whether there is, if so, going to step 2；If it is not, then
Go to step 3；
The disaggregated model of step 2, real-time processing module calling model training module, classifies to user message, if rubbish
Message, then export rubbish message as a result, this user message processing terminate；If normal messages then go to step 3；
Step 3, real-time processing module differentiate whether judge keyword to user message by LP algorithm and key word analysis
It is abnormal, if it is not, then export normal messages as a result, this user message processing terminate；If so, will be used by key word analysis
Family message is labeled as rubbish message, feeds back to rubbish message discovery by the type of LP algorithm tag user message, and by result
Module；
Step 4, rubbish message discovery module are handled user message and its type by PU learning algorithm, are obtained
New sample message simultaneously updates sample database；Model training module is trained using the sample message of sample database and updates classification mould
Type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289668.5A CN110059189B (en) | 2019-04-11 | 2019-04-11 | Game platform message classification system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289668.5A CN110059189B (en) | 2019-04-11 | 2019-04-11 | Game platform message classification system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110059189A true CN110059189A (en) | 2019-07-26 |
| CN110059189B CN110059189B (en) | 2021-03-05 |
Family
ID=67317608
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910289668.5A Active CN110059189B (en) | 2019-04-11 | 2019-04-11 | Game platform message classification system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059189B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111198947A (en) * | 2020-01-06 | 2020-05-26 | 南京中新赛克科技有限责任公司 | Convolutional neural network fraud short message classification method and system based on naive Bayes optimization |
| WO2021042828A1 (en) * | 2019-09-04 | 2021-03-11 | 华为技术有限公司 | Neural network model compression method and apparatus, and storage medium and chip |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1801854A (en) * | 2004-12-21 | 2006-07-12 | 朗迅科技公司 | Detection of unwanted messages (spam) |
| CN101106748A (en) * | 2006-07-11 | 2008-01-16 | 华为技术有限公司 | A content filtering system, device and method for mobile network |
| CN101317376A (en) * | 2006-07-11 | 2008-12-03 | 华为技术有限公司 | Method, device and system for contents filtering |
| CN101415161A (en) * | 2008-11-26 | 2009-04-22 | 深圳华为通信技术有限公司 | Method and apparatus for filtrating rubbish short message |
| CN101516071A (en) * | 2008-02-18 | 2009-08-26 | 中国移动通信集团重庆有限公司 | Method for classifying junk short messages |
| US7836061B1 (en) * | 2007-12-29 | 2010-11-16 | Kaspersky Lab, Zao | Method and system for classifying electronic text messages and spam messages |
| CN103312585A (en) * | 2012-03-08 | 2013-09-18 | 中兴通讯股份有限公司 | Processing method and system of spam message |
| CN103634473A (en) * | 2013-12-05 | 2014-03-12 | 南京理工大学连云港研究院 | Naive Bayesian classification based mobile phone spam short message filtering method and system |
| US20150363368A1 (en) * | 2013-10-24 | 2015-12-17 | Vmware, Inc. | Transforming HTML Forms into Mobile Native Forms |
| CN105389379A (en) * | 2015-11-20 | 2016-03-09 | 重庆邮电大学 | Rubbish article classification method based on distributed feature representation of text |
| CN105488029A (en) * | 2015-11-30 | 2016-04-13 | 西安闻泰电子科技有限公司 | KNN based evidence taking method for instant communication tool of intelligent mobile phone |
| US20160232231A1 (en) * | 2015-02-11 | 2016-08-11 | Hung Dang Viet | System and method for document and/or message document and/or message content suggestion, user rating and user reward |
| CN106202177A (en) * | 2016-06-27 | 2016-12-07 | 腾讯科技（深圳）有限公司 | A kind of file classification method and device |
| CN106937262A (en) * | 2015-12-29 | 2017-07-07 | 中国移动通信集团公司 | A kind of message treatment method, apparatus and system |
| CN108463815A (en) * | 2016-01-07 | 2018-08-28 | Mz知识产权控股有限责任公司 | The name Entity recognition of chat data |
| US20190068541A1 (en) * | 2015-11-30 | 2019-02-28 | International Business Machines Corporation | Determining intended electronic message recipients via linguistic profiles |
| WO2019040279A1 (en) * | 2017-08-22 | 2019-02-28 | Gali Health, Inc. | Personalized digital health system using temporal models |
| US10346448B2 (en) * | 2016-07-13 | 2019-07-09 | Google Llc | System and method for classifying an alphanumeric candidate identified in an email message |
-
2019
- 2019-04-11 CN CN201910289668.5A patent/CN110059189B/en active Active
Patent Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1801854A (en) * | 2004-12-21 | 2006-07-12 | 朗迅科技公司 | Detection of unwanted messages (spam) |
| CN101106748A (en) * | 2006-07-11 | 2008-01-16 | 华为技术有限公司 | A content filtering system, device and method for mobile network |
| CN101317376A (en) * | 2006-07-11 | 2008-12-03 | 华为技术有限公司 | Method, device and system for contents filtering |
| US7836061B1 (en) * | 2007-12-29 | 2010-11-16 | Kaspersky Lab, Zao | Method and system for classifying electronic text messages and spam messages |
| CN101516071A (en) * | 2008-02-18 | 2009-08-26 | 中国移动通信集团重庆有限公司 | Method for classifying junk short messages |
| CN101415161A (en) * | 2008-11-26 | 2009-04-22 | 深圳华为通信技术有限公司 | Method and apparatus for filtrating rubbish short message |
| CN103312585A (en) * | 2012-03-08 | 2013-09-18 | 中兴通讯股份有限公司 | Processing method and system of spam message |
| US20150363368A1 (en) * | 2013-10-24 | 2015-12-17 | Vmware, Inc. | Transforming HTML Forms into Mobile Native Forms |
| CN103634473A (en) * | 2013-12-05 | 2014-03-12 | 南京理工大学连云港研究院 | Naive Bayesian classification based mobile phone spam short message filtering method and system |
| US20160232231A1 (en) * | 2015-02-11 | 2016-08-11 | Hung Dang Viet | System and method for document and/or message document and/or message content suggestion, user rating and user reward |
| CN105389379A (en) * | 2015-11-20 | 2016-03-09 | 重庆邮电大学 | Rubbish article classification method based on distributed feature representation of text |
| CN105488029A (en) * | 2015-11-30 | 2016-04-13 | 西安闻泰电子科技有限公司 | KNN based evidence taking method for instant communication tool of intelligent mobile phone |
| US20190068541A1 (en) * | 2015-11-30 | 2019-02-28 | International Business Machines Corporation | Determining intended electronic message recipients via linguistic profiles |
| CN106937262A (en) * | 2015-12-29 | 2017-07-07 | 中国移动通信集团公司 | A kind of message treatment method, apparatus and system |
| CN108463815A (en) * | 2016-01-07 | 2018-08-28 | Mz知识产权控股有限责任公司 | The name Entity recognition of chat data |
| CN106202177A (en) * | 2016-06-27 | 2016-12-07 | 腾讯科技（深圳）有限公司 | A kind of file classification method and device |
| US10346448B2 (en) * | 2016-07-13 | 2019-07-09 | Google Llc | System and method for classifying an alphanumeric candidate identified in an email message |
| WO2019040279A1 (en) * | 2017-08-22 | 2019-02-28 | Gali Health, Inc. | Personalized digital health system using temporal models |
Non-Patent Citations (2)
| Title |
|---|
| 张蕾: ""机器学习在网络空间安全研究中的应用"", 《计算机学报》 * |
| 郭浩: ""多特征微博垃圾互粉检测方法"", 《中国科技论文》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021042828A1 (en) * | 2019-09-04 | 2021-03-11 | 华为技术有限公司 | Neural network model compression method and apparatus, and storage medium and chip |
| CN111198947A (en) * | 2020-01-06 | 2020-05-26 | 南京中新赛克科技有限责任公司 | Convolutional neural network fraud short message classification method and system based on naive Bayes optimization |
| CN111198947B (en) * | 2020-01-06 | 2024-02-13 | 南京中新赛克科技有限责任公司 | Convolutional neural network fraud short message classification method and system based on naive Bayes optimization |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110059189B (en) | 2021-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110826337B (en) | Short text semantic training model acquisition method and similarity matching algorithm | |
| CN106095928B (en) | A kind of event type recognition methods and device | |
| CN107515873B (en) | Junk information identification method and equipment | |
| CN112613552B (en) | Convolutional neural network emotion image classification method combined with emotion type attention loss | |
| CN109871885A (en) | A kind of plants identification method based on deep learning and Plant Taxonomy | |
| CN109034194A (en) | Transaction swindling behavior depth detection method based on feature differentiation | |
| CN106709370B (en) | A kind of long word recognition method and system based on content of text | |
| CN109299271A (en) | Training sample generation, text data, public sentiment event category method and relevant device | |
| CN109872162A (en) | A kind of air control classifying identification method and system handling customer complaint information | |
| CN111461216A (en) | Case risk identification method based on machine learning | |
| CN103854063A (en) | Internet open information-based event occurrence risk prediction and early-warning method | |
| CN110059189A (en) | A kind of categorizing system and method for gaming platform message | |
| CN113376516A (en) | Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning | |
| CN110909542B (en) | Intelligent semantic serial-parallel analysis method and system | |
| CN104156447A (en) | Intelligent social platform advertisement early warning and handling method | |
| CN113327140A (en) | Video advertisement putting effect intelligent analysis management system based on big data analysis | |
| CN107357895A (en) | A kind of processing method of the text representation based on bag of words | |
| CN110110087A (en) | A kind of Feature Engineering method for Law Text classification based on two classifiers | |
| CN109299251A (en) | A kind of abnormal refuse messages recognition methods and system based on deep learning algorithm | |
| CN116541755A (en) | Financial behavior pattern analysis and prediction method based on time sequence diagram representation learning | |
| CN111047428A (en) | Bank high-risk fraud client identification method based on small amount of fraud samples | |
| CN115600602B (en) | Method, system and terminal device for extracting key elements of long text | |
| CN110362828B (en) | Network information risk identification method and system | |
| CN114912026B (en) | Network public opinion monitoring analysis processing method, equipment and computer storage medium | |
| CN109858901A (en) | Mobile subscriber based on mobile phone signaling big data pays recognition methods |
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 |