CN109165350A - A kind of information recommendation method and system based on deep knowledge perception - Google Patents
A kind of information recommendation method and system based on deep knowledge perception Download PDFInfo
- Publication number
- CN109165350A CN109165350A CN201810967174.3A CN201810967174A CN109165350A CN 109165350 A CN109165350 A CN 109165350A CN 201810967174 A CN201810967174 A CN 201810967174A CN 109165350 A CN109165350 A CN 109165350A
- Authority
- CN
- China
- Prior art keywords
- information
- entity
- feature
- user
- knowledge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The present invention discloses a kind of information recommendation method and system based on deep knowledge perception, comprising: obtains the entity set of user's history click information；Each of entity set to history click information entity, lookup and its entity with knowledge connection in the knowledge mapping pre-established, obtains entity vector set；To each of entity vector set entity, corresponding Feature Words are searched in knowledge mapping, obtain term vector collection；Feature extraction is carried out to the default feature of term vector collection and history click information, obtains historical information feature set；Recommend one or more Candidate Recommendation information according to historical information feature set；User is calculated for the click probability of each Candidate Recommendation information；Probability be will click on according to sorting from large to small, generate information recommendation list.The content of the content of information text semantic level and knowledge level can be blended carry out feature extraction, to improve the accuracy of recommendation information by technical solution provided by the invention.
Description
Technical field
The present invention relates to depth learning technology field more particularly to a kind of information recommendation methods based on deep knowledge perception
And system.
Background technique
With the development of internet technology, there is different degrees of information overload phenomenon in all kinds of websites and APP how
Comforming and selecting suitable commending contents in multi information to user is each website and APP developer's problem encountered.And it solves
It is that user carries out personalized commending contents that certainly the method for this problem, which is exactly using recommender system, i.e., previous by analysis user
Click on content feature, obtain the potential interest and focus of the user, by the interested content push of user give the user,
To improve website and the income of APP.
Content-based recommendation be based on user using website or APP when the content made of click on content similar push away
It recommends.Content information is usually with the shape of text or class text (for example, recording mode that image, voice etc. can be converted into text information)
Formula exists.The existing information recommendation based on content is only learnt from the semantic level of text, without making full use of information
Between knowledge level connection.For the information brief particularly with some content of text, hot spot timeliness is high, only in semanteme
Level obtains information characteristics, can not accurately identify the knowledge entity of text behind, to can not recommend to be more in line with it to user
The information of reading interest.
Summary of the invention
The present invention is intended to provide a kind of information recommendation method and system based on deep knowledge perception, it can be by information text
The content of semantic level and the content of knowledge level blend carry out feature extraction, to improve the accuracy of recommendation information.
In order to achieve the above objectives, The technical solution adopted by the invention is as follows:
A kind of information recommendation method based on deep knowledge perception, comprising: obtain the entity set of user's history click information；
Each of entity set to history click information entity is searched to have with it in the knowledge mapping pre-established and be known
Know associated entity, obtains entity vector set；To each of entity vector set entity, looked into the knowledge mapping
Corresponding Feature Words are looked for, term vector collection is obtained；The knowledge mapping is made of entity and Feature Words, each entity corresponding one
A features above word, each entity have more than one knowledge connection entity；Letter is clicked to the term vector collection and the history
The default feature of breath carries out feature extraction, obtains historical information feature set；According to the historical information feature set recommend one with
Upper Candidate Recommendation information；User is calculated for the click probability of Candidate Recommendation information described in each；The click probability is pressed
According to sorting from large to small, information recommendation list is generated.
Preferably, described to calculate user for the method for the click probability of Candidate Recommendation information described in each are as follows: to obtain
The feature term vector of every Candidate Recommendation information；The Candidate Recommendation information is generated using the attention model pre-established
Feature term vector weight；The training set of the attention model is the Feature Words vector set of user's history click information；Root
According to the weight and the user information that obtains in advance of the feature term vector of the Candidate Recommendation information, using the shot and long term pre-established
Memory network calculates user for the click probability of Candidate Recommendation information described in each.
Preferably, the user information obtained in advance includes: user's head portrait, age of user, user's gender etc..
Preferably, the entity set for obtaining user's history click information includes: to divide the history click information
Word processing, filters stop words, obtains the history click information word set；Know described in identification from the history click information word set
Know the entity recorded in map, obtains primary entities collection；Using entity link technology to each of described primary entities collection
Entity disambiguation obtains the entity set of the user's history click information.
Preferably, the default feature of the history click information includes: the history click information word set, and the user goes through
The entity set of history click information corresponding feature word set in the knowledge mapping.
Preferably, the default feature to the term vector collection and the history click information carries out feature extraction, obtains
Taking historical information feature set includes: to generate the first word using word embedding technology to the history click information word set
Vector matrix；Entity set corresponding feature word set, use in the knowledge mapping to the user's history click information
Word embedding technology generates the second term vector matrix；To the entity set in the knowledge mapping step knowledge connection
Entity set in the knowledge mapping corresponding Feature Words vector set, using word embedding technology generate third word to
Moment matrix；Using the first term vector matrix, the second term vector matrix, third term vector matrix as the convolution mind pre-established
Three channels through network input layer obtain the historical information feature set after convolutional neural networks calculating.
A kind of information recommendation system based on deep knowledge perception, comprising: knowledge mapping, the knowledge mapping by entity and
Feature Words composition, each entity correspond to more than one Feature Words, and each entity has more than one knowledge connection entity；Entity set
Module is obtained, for obtaining the entity set of user's history click information；First searching module, for the history click information
Each of entity set entity, searched in the knowledge mapping and its entity with knowledge connection, obtain entity to
Quantity set；Second searching module, for each of entity vector set entity, searched in the knowledge mapping and its
Corresponding Feature Words obtain term vector collection；Characteristic extracting module, for the term vector collection and the history click information
Default feature carries out feature extraction, obtains historical information feature set；Recommending module, for being pushed away according to the historical information feature set
Recommend one or more Candidate Recommendation information；Computing module, for calculating click of the user for Candidate Recommendation information described in each
Probability；Recommendation list generation module, for the click probability according to sorting from large to small, to be generated information recommendation list.
Preferably, the computing module includes: that feature term vector obtains module, for obtaining the every Candidate Recommendation letter
The feature term vector of breath；Attention model, the weight of the feature term vector for generating the Candidate Recommendation information；The attention
The training set of power model is the Feature Words vector set of user's history click information；Shot and long term memory network, for according to the time
The user information selecting the weight of the feature term vector of recommendation information and obtaining in advance calculates user and candidate described in each is pushed away
Recommend the click probability of information.
Preferably, it includes: word segmentation module that the entity set, which obtains module, for segmenting to the history click information
Processing filters stop words, obtains the history click information word set；Entity recognition module is used for from the history click information
The entity recorded in the knowledge mapping is identified in word set, obtains primary entities collection；Entity link module, for described original
The entity disambiguation of each of entity set obtains the entity set of the user's history click information.
Preferably, the default feature of the history click information includes: the history click information word set, and the user goes through
The entity set of history click information corresponding feature word set in the knowledge mapping；The characteristic extracting module includes: word
Embedding module, for generating the first term vector matrix to the history click information word set；The user's history is clicked
The entity set of information corresponding feature word set in the knowledge mapping generates the second term vector matrix；To the entity set in institute
It states the entity set of a step knowledge connection in the knowledge mapping corresponding Feature Words vector set in the knowledge mapping and generates third word
Vector matrix；Convolutional neural networks are used for the first term vector matrix, the second term vector matrix, third term vector matrix
As three channels of the convolutional neural networks input layer, calculates and obtain the historical information feature set.
Information recommendation method and system provided in an embodiment of the present invention based on deep knowledge perception, is obtaining user's history
After the entity set of click information, using the knowledge mapping that pre-establishes obtain with above-mentioned entity set have the entity of knowledge connection to
Quantity set, and then obtain the corresponding term vector collection of entity vector set and entity set step knowledge connection in the knowledge mapping
The corresponding Feature Words vector set of entity set, after carrying out feature extraction to all term vector collection, can recommend semantic level and to know
The information that knowledge level blends is to user.Compared with the existing recommended technology for only considering information text semantic level, recommendation
Information is more accurate, more meets the reading interest of user.Meanwhile the present invention also counts the click probability of Candidate Recommendation information
It calculates, and will click on probability and sort from large to small, preferred display clicks the big information of probability.In conclusion skill provided by the invention
The content of the content of information text semantic level and knowledge level can be blended carry out feature extraction, to mention by art scheme
The accuracy of high recommendation information, and then improve user's clicking rate.
Detailed description of the invention
Fig. 1 is the method flow diagram of the embodiment of the present invention；
Fig. 2 is the system construction drawing one of the embodiment of the present invention；
Fig. 3 is the system construction drawing two of the embodiment of the present invention；
Fig. 4 is the method flow diagram that the entity set of user's history click information is obtained in the embodiment of the present invention；
Fig. 5 is that user is calculated in the embodiment of the present invention for the method for the click probability of Candidate Recommendation information described in each
Flow chart；
Fig. 6 is the structural schematic diagram of knowledge mapping in the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing, to the present invention into
Row is further described.
Fig. 1 is the method flow diagram of the embodiment of the present invention, comprising the following steps:
Step 101, the entity set of user's history click information is obtained；
In the present embodiment, it is described obtain user's history click information entity set the following steps are included:
Step 1011, word segmentation processing is carried out to the history click information, filters stop words, obtained the history and click letter
Cease word set；
Step 1012, the entity recorded in the knowledge mapping is identified from the history click information word set, is obtained former
Beginning entity set；
Step 1013, the entity disambiguation of each of the primary entities collection is obtained using entity link technology
The entity set of the user's history click information.Entity link technology in the present invention is the entity link skill in knowledge mapping
Art, for example, entity " Shanghai " is associated with out in knowledge mapping using entity link technology comprising " evil spirit is all " in information, with " on
Sea " replacement " evil spirit is all ", with disambiguation.
Step 102, to each of the entity set of history click information entity, in the knowledge mapping pre-established
Middle lookup and its entity with knowledge connection obtain entity vector set；
In the present embodiment, the knowledge mapping is made of entity and Feature Words, and each entity corresponds to more than one Feature Words,
Each entity has more than one knowledge connection entity, as shown in Figure 3.For example, some reality in the entity set of history click information
Body is " the Forbidden City ", then having the entity of knowledge connection with " the Forbidden City " includes: " Ming Dynasty ", " Beijing ", " imperial palace ", " world five
Big palace " etc., the entity vector of generation are [Ming Dynasty, Beijing, imperial palace, the big palace ... ... in the world five].Institute in history click information
Some entities obtain entity vector set after above-mentioned knowledge connection.
Step 103, it to each of entity vector set entity, is searched in the knowledge mapping corresponding
Feature Words obtain term vector collection；
Step 104, feature extraction is carried out to the default feature of the term vector collection and the history click information, acquisition is gone through
History information characteristics collection；
In the present embodiment, the default feature of the history click information includes: the history point obtained in a step 101
Information word set is hit, the entity set of the user's history click information obtained in a step 101 is corresponding in the knowledge mapping
Feature word set.Then the concrete methods of realizing of step 104 includes: to the history click information word set, using word
Embedding technology generates the first term vector matrix；To the entity set of the user's history click information in the knowledge mapping
In corresponding feature word set, using word embedding technology generate the second term vector matrix；To the entity set described
The entity set of a step knowledge connection corresponding Feature Words vector set in the knowledge mapping in knowledge mapping, using word
Embedding technology generates third term vector matrix；By the first term vector matrix, the second term vector matrix, third word to
Three channels of the moment matrix as the convolutional neural networks input layer pre-established obtain after convolutional neural networks calculating
The historical information feature set.Above-mentioned first term vector matrix, the second term vector matrix, third term vector matrix are similar to image
Tri- channels RGB in identification, are inputted the convolutional neural networks pre-established, obtain historical information feature set.Above-mentioned side
In method, each history click information can all obtain corresponding historical information feature set, and the method spliced using vector can obtain
The feature set of all history click informations.
Word embedding is one in NLP (Neuro-Linguistic Programming, neural LISP program LISP)
Group language model (language modeling) and feature learning technology (feature learning techniques) it is total
Claim, these technologies can in vocabulary word or phrase (words or phrases) be mapped to the vector being made of real number
On.In the present embodiment, it may be selected during word embedding with common word2vec method.
Step 105, one or more Candidate Recommendation information is recommended according to the historical information feature set；
Step 106, user is calculated for the click probability of Candidate Recommendation information described in each；
In the present embodiment, step 106 method particularly includes:
Step 1061, the feature term vector of every Candidate Recommendation information is obtained；
Step 1062, the power of the feature term vector of the Candidate Recommendation information is generated using the attention model pre-established
Weight；The training set of the attention model is the Feature Words vector set of user's history click information；Attention model
(Attention Model) is the model that human brain attention is simulated in deep learning, is used to be weighted target data change
Change.By trained attention model, the weight for generating feature term vector can be calculated.
Step 1063, the user information obtained according to the weight of the feature term vector of the Candidate Recommendation information and in advance,
User is calculated for the click probability of Candidate Recommendation information described in each using the shot and long term memory network pre-established.It is described
The user information obtained in advance includes: user's head portrait, age of user, user's gender etc..
In this step, using above-mentioned user information as the long-term characteristic of user, the feature for the information that user is clicked in the recent period
Every Candidate Recommendation information is calculated in conjunction with the weight of the feature term vector of calculated recommendation information as Short-term characteristic
Click probability.
For the new user of a website or APP, there is no history click information, the heat of certain time can be used at this time
Door information replaces the history click information of the user.Therefore, even new user, attention model can also be used and generate candidate
The weight of the feature term vector of recommendation information, and then calculate and click probability.
Step 107, the click probability is generated into information recommendation list according to sorting from large to small.Information recommendation is arranged
Preceding n information in table is presented to user.
Invention additionally discloses it is a kind of based on deep knowledge perception message recommender system, structural schematic diagram as shown in Fig. 2,
It include: knowledge mapping, the knowledge mapping is made of entity and Feature Words, and each entity corresponds to more than one Feature Words, each
Entity has more than one knowledge connection entity；Entity set obtains module, for obtaining the entity set of user's history click information；
First searching module is looked into the knowledge mapping for each of entity set to history click information entity
The entity with it with knowledge connection is looked for, entity vector set is obtained；Second searching module, for in the entity vector set
Each entity searches corresponding Feature Words in the knowledge mapping, obtains term vector collection；Characteristic extracting module is used
Feature extraction is carried out in the default feature to the term vector collection and the history click information, obtains historical information feature set；
Recommending module, for recommending one or more Candidate Recommendation information according to the historical information feature set；Computing module, for calculating
Click probability of the user for Candidate Recommendation information described in each；Recommendation list generation module is used for the click probability
According to sorting from large to small, information recommendation list is generated.
In the present embodiment, as shown in figure 3, the computing module includes: that feature term vector obtains module, for obtaining every
The feature term vector of the Candidate Recommendation information；Attention model, for generating the feature term vector of the Candidate Recommendation information
Weight；The training set of the attention model is the Feature Words vector set of user's history click information；Shot and long term memory network,
User information for obtaining according to the weight of the feature term vector of the Candidate Recommendation information and in advance, calculates user for every
The click probability of the one Candidate Recommendation information.
In the present embodiment, as shown in figure 3, it includes: word segmentation module that the entity set, which obtains module, for the history point
It hits information and carries out word segmentation processing, filter stop words, obtain the history click information word set；Entity recognition module is used for from institute
It states and identifies the entity recorded in the knowledge mapping in history click information word set, obtain primary entities collection；Entity link module,
For obtaining the entity set of the user's history click information to the entity disambiguation of each of the primary entities collection.
In the present embodiment, the default feature of the history click information includes: the history click information word set, the use
The entity set of family history click information corresponding Feature Words vector set in the knowledge mapping, the entity set is in the knowledge
The entity set of a step knowledge connection corresponding Feature Words vector set in the knowledge mapping in map；The characteristic extracting module
It include: word embedding module, for generating the first term vector matrix to the history click information word set；To the use
The entity set of family history click information corresponding feature word set in the knowledge mapping generates the second term vector matrix；To described
The entity set of entity set step knowledge connection in the knowledge mapping corresponding Feature Words moment of a vector in the knowledge mapping
Battle array；Convolutional neural networks, for using the first term vector matrix, the second term vector matrix, third term vector matrix as institute
Three channels of convolutional neural networks input layer are stated, calculates and obtains the historical information feature set.
Information recommendation method and system provided in an embodiment of the present invention based on deep knowledge perception, is obtaining user's history
After the entity set of click information, using the knowledge mapping that pre-establishes obtain with above-mentioned entity set have the entity of knowledge connection to
Quantity set, and then the corresponding term vector collection of entity vector set is obtained, after term vector collection progress feature extraction, semanteme can be recommended
The information that level and knowledge level blend is to user.With the existing recommended technology phase for only considering information text semantic level
Than the information of recommendation is more accurate, more meets the reading interest of user.Meanwhile click of the present invention also to Candidate Recommendation information
Probability is calculated, and be will click on probability and sorted from large to small, and preferred display clicks the big information of probability.In conclusion this hair
The content of the content of information text semantic level and knowledge level can be blended progress feature and mentioned by the technical solution of bright offer
It takes, to improve the accuracy of recommendation information, and then improves user's clicking rate.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.
Claims (10)
1. a kind of information recommendation method based on deep knowledge perception characterized by comprising
Obtain the entity set of user's history click information；
Each of entity set to history click information entity, searches in the knowledge mapping pre-established and has with it
There is the entity of knowledge connection, obtains entity vector set；
To each of entity vector set entity, corresponding Feature Words are searched in the knowledge mapping, are obtained
Term vector collection；The knowledge mapping is made of entity and Feature Words, and each entity corresponds to more than one Feature Words, each entity tool
There is more than one knowledge connection entity；
Feature extraction is carried out to the default feature of the term vector collection and the history click information, obtains historical information feature
Collection；
Recommend one or more Candidate Recommendation information according to the historical information feature set；
User is calculated for the click probability of Candidate Recommendation information described in each；
By the click probability according to sorting from large to small, information recommendation list is generated.
2. the information recommendation method according to claim 1 based on deep knowledge perception, which is characterized in that the calculating is used
Method of the family for the click probability of Candidate Recommendation information described in each are as follows:
Obtain the feature term vector of every Candidate Recommendation information；
The weight of the feature term vector of the Candidate Recommendation information is generated using the attention model pre-established；The attention
The training set of model is the Feature Words vector set of user's history click information；
The user information obtained according to the weight of the feature term vector of the Candidate Recommendation information and in advance, using what is pre-established
Shot and long term memory network calculates user for the click probability of Candidate Recommendation information described in each.
3. the information recommendation method according to claim 2 based on deep knowledge perception, which is characterized in that described to obtain in advance
The user information taken includes: user's head portrait, age of user, user's gender.
4. the information recommendation method according to claim 1 based on deep knowledge perception, which is characterized in that the acquisition is used
The entity set of family history click information includes:
Word segmentation processing is carried out to the history click information, stop words is filtered, obtains the history click information word set；
The entity recorded in the knowledge mapping is identified from the history click information word set, obtains primary entities collection；
Using entity link technology to the entity disambiguation of each of the primary entities collection, the user's history point is obtained
Hit the entity set of information.
5. the information recommendation method according to claim 4 based on deep knowledge perception, which is characterized in that the history point
The default feature for hitting information includes: the history click information word set, and the entity set of the user's history click information is described
Corresponding feature word set in knowledge mapping.
6. the information recommendation method according to claim 5 based on deep knowledge perception, which is characterized in that described to described
The default feature of term vector collection and the history click information carries out feature extraction, obtains historical information feature set and includes:
To the history click information word set, the first term vector matrix is generated using word embedding technology；
Vector set corresponding feature word set in the knowledge mapping is corresponded to the entity set of the user's history click information, is adopted
The second term vector matrix is generated with word embedding technology；
To the step knowledge connection entity set of the entity set of the user's history click information in knowledge mapping in the knowledge
Corresponding feature word set in map generates third term vector matrix using word embedding technology；
Using the first term vector matrix, the second term vector matrix, third term vector matrix as the convolutional Neural pre-established
Three channels of network input layer obtain the historical information feature set after convolutional neural networks calculating.
7. a kind of information recommendation system based on deep knowledge perception characterized by comprising
Knowledge mapping, the knowledge mapping are made of entity and Feature Words, and each entity corresponds to more than one Feature Words, Mei Geshi
Body has more than one knowledge connection entity；
Entity set obtains module, for obtaining the entity set of user's history click information；
First searching module, for each of entity set to history click information entity, in the knowledge mapping
Middle lookup and its entity with knowledge connection obtain entity vector set；
Second searching module, for each of entity vector set entity, searched in the knowledge mapping and its
Corresponding Feature Words obtain term vector collection；
Characteristic extracting module carries out feature extraction for the default feature to the term vector collection and the history click information,
Obtain historical information feature set；
Recommending module, for recommending one or more Candidate Recommendation information according to the historical information feature set；
Computing module, for calculating user for the click probability of Candidate Recommendation information described in each；
Recommendation list generation module, for the click probability according to sorting from large to small, to be generated information recommendation list.
8. the information recommendation system according to claim 7 based on deep knowledge perception, which is characterized in that the calculating mould
Block includes:
Feature term vector obtains module, for obtaining the feature term vector of every Candidate Recommendation information；
Attention model, the weight of the feature term vector for generating the Candidate Recommendation information；The instruction of the attention model
Practice the Feature Words vector set integrated as user's history click information；
Shot and long term memory network, the user for obtaining according to the weight of the feature term vector of the Candidate Recommendation information and in advance
Information calculates user for the click probability of Candidate Recommendation information described in each.
9. the information recommendation system according to claim 7 based on deep knowledge perception, which is characterized in that the entity set
Obtaining module includes:
Word segmentation module filters stop words, obtains the history and click letter for carrying out word segmentation processing to the history click information
Cease word set；
Entity recognition module is obtained for identifying the entity recorded in the knowledge mapping from the history click information word set
Take primary entities collection；
Entity link module, for obtaining the user's history to the entity disambiguation of each of the primary entities collection
The entity set of click information.
10. the information recommendation system according to claim 9 based on deep knowledge perception, which is characterized in that the history
The default feature of click information includes: the history click information word set, and the entity set of the user's history click information is in institute
State corresponding feature word set in knowledge mapping；The characteristic extracting module includes:
Word embedding module, for generating the first term vector matrix to the history click information word set；To the use
The entity set of family history click information corresponding Feature Words vector set in the knowledge mapping generates the second term vector matrix；It is right
The entity set of entity set step knowledge connection in the knowledge mapping in the knowledge mapping corresponding Feature Words to
Quantity set generates third term vector matrix；
Convolutional neural networks, for using the first term vector matrix, the second term vector matrix, third term vector matrix as institute
Three channels of convolutional neural networks input layer are stated, calculates and obtains the historical information feature set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810967174.3A CN109165350A (en) | 2018-08-23 | 2018-08-23 | A kind of information recommendation method and system based on deep knowledge perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810967174.3A CN109165350A (en) | 2018-08-23 | 2018-08-23 | A kind of information recommendation method and system based on deep knowledge perception |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109165350A true CN109165350A (en) | 2019-01-08 |
Family
ID=64896479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810967174.3A Pending CN109165350A (en) | 2018-08-23 | 2018-08-23 | A kind of information recommendation method and system based on deep knowledge perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109165350A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105869016A (en) * | 2016-03-28 | 2016-08-17 | 天津中科智能识别产业技术研究院有限公司 | Method for estimating click through rate based on convolution neural network |
| CN109858528A (en) * | 2019-01-10 | 2019-06-07 | 平安科技（深圳）有限公司 | Recommender system training method, device, computer equipment and storage medium |
| CN110149541A (en) * | 2019-04-23 | 2019-08-20 | 腾讯科技（深圳）有限公司 | Video recommendation method, device, computer equipment and storage medium |
| CN110263179A (en) * | 2019-06-12 | 2019-09-20 | 湖南酷得网络科技有限公司 | Learning path method for pushing, device, computer equipment and storage medium |
| CN110264301A (en) * | 2019-05-10 | 2019-09-20 | 拉扎斯网络科技（上海）有限公司 | Recommended method, device, electronic equipment and non-volatile memory medium |
| CN110288436A (en) * | 2019-06-19 | 2019-09-27 | 桂林电子科技大学 | A kind of personalized recommending scenery spot method based on the modeling of tourist's preference |
| CN110634047A (en) * | 2019-09-05 | 2019-12-31 | 北京无限光场科技有限公司 | Method and device for recommending house resources, electronic equipment and storage medium |
| CN110888990A (en) * | 2019-11-22 | 2020-03-17 | 深圳前海微众银行股份有限公司 | Text recommendation method, device, equipment and medium |
| CN111061856A (en) * | 2019-06-06 | 2020-04-24 | 北京理工大学 | Knowledge perception-based news recommendation method |
| CN111147375A (en) * | 2019-12-31 | 2020-05-12 | 秒针信息技术有限公司 | Network operation method, device, computer equipment and medium |
| CN111190968A (en) * | 2019-12-16 | 2020-05-22 | 北京航天智造科技发展有限公司 | Data preprocessing and content recommendation method based on knowledge graph |
| CN111460808A (en) * | 2020-03-23 | 2020-07-28 | 腾讯科技（深圳）有限公司 | Synonymous text recognition and content recommendation method and device and electronic equipment |
| CN112163149A (en) * | 2020-09-16 | 2021-01-01 | 北京明略昭辉科技有限公司 | Method and device for recommending messages |
| CN112201348A (en) * | 2020-10-28 | 2021-01-08 | 浙江大学 | Multi-center clinical data set adapting device based on knowledge perception |
| CN112214685A (en) * | 2020-09-27 | 2021-01-12 | 电子科技大学 | Knowledge graph-based personalized recommendation method |
| CN112528010A (en) * | 2020-12-15 | 2021-03-19 | 建信金融科技有限责任公司 | Knowledge recommendation method and device, computer equipment and readable storage medium |
| CN113449176A (en) * | 2020-03-24 | 2021-09-28 | 华为技术有限公司 | Recommendation method and device based on knowledge graph |
| CN113961823A (en) * | 2021-12-17 | 2022-01-21 | 江西中业智能科技有限公司 | News recommendation method, system, storage medium and equipment |
| CN112163149B (en) * | 2020-09-16 | 2024-05-03 | 北京明略昭辉科技有限公司 | Method and device for recommending message |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104572734A (en) * | 2013-10-23 | 2015-04-29 | 腾讯科技（深圳）有限公司 | Question recommendation method, device and system |
| CN106570144A (en) * | 2016-02-05 | 2017-04-19 | 中科鼎富（北京）科技发展有限公司 | Method and apparatus for recommending information |
| CN107577763A (en) * | 2017-09-04 | 2018-01-12 | 北京京东尚科信息技术有限公司 | Search method and device |
| CN108345702A (en) * | 2018-04-10 | 2018-07-31 | 北京百度网讯科技有限公司 | Entity recommends method and apparatus |
-
2018
- 2018-08-23 CN CN201810967174.3A patent/CN109165350A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104572734A (en) * | 2013-10-23 | 2015-04-29 | 腾讯科技（深圳）有限公司 | Question recommendation method, device and system |
| CN106570144A (en) * | 2016-02-05 | 2017-04-19 | 中科鼎富（北京）科技发展有限公司 | Method and apparatus for recommending information |
| CN107577763A (en) * | 2017-09-04 | 2018-01-12 | 北京京东尚科信息技术有限公司 | Search method and device |
| CN108345702A (en) * | 2018-04-10 | 2018-07-31 | 北京百度网讯科技有限公司 | Entity recommends method and apparatus |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105869016A (en) * | 2016-03-28 | 2016-08-17 | 天津中科智能识别产业技术研究院有限公司 | Method for estimating click through rate based on convolution neural network |
| CN109858528A (en) * | 2019-01-10 | 2019-06-07 | 平安科技（深圳）有限公司 | Recommender system training method, device, computer equipment and storage medium |
| CN110149541B (en) * | 2019-04-23 | 2021-08-03 | 腾讯科技（深圳）有限公司 | Video recommendation method and device, computer equipment and storage medium |
| CN110149541A (en) * | 2019-04-23 | 2019-08-20 | 腾讯科技（深圳）有限公司 | Video recommendation method, device, computer equipment and storage medium |
| US11540019B2 (en) | 2019-04-23 | 2022-12-27 | Tencent Technology (Shenzhen) Company Limited | Video recommendation method and device, computer device and storage medium |
| CN110264301A (en) * | 2019-05-10 | 2019-09-20 | 拉扎斯网络科技（上海）有限公司 | Recommended method, device, electronic equipment and non-volatile memory medium |
| CN111061856A (en) * | 2019-06-06 | 2020-04-24 | 北京理工大学 | Knowledge perception-based news recommendation method |
| CN111061856B (en) * | 2019-06-06 | 2022-05-27 | 北京理工大学 | Knowledge perception-based news recommendation method |
| CN110263179A (en) * | 2019-06-12 | 2019-09-20 | 湖南酷得网络科技有限公司 | Learning path method for pushing, device, computer equipment and storage medium |
| CN110288436A (en) * | 2019-06-19 | 2019-09-27 | 桂林电子科技大学 | A kind of personalized recommending scenery spot method based on the modeling of tourist's preference |
| CN110634047A (en) * | 2019-09-05 | 2019-12-31 | 北京无限光场科技有限公司 | Method and device for recommending house resources, electronic equipment and storage medium |
| CN110634047B (en) * | 2019-09-05 | 2022-04-22 | 北京无限光场科技有限公司 | Method and device for recommending house resources, electronic equipment and storage medium |
| CN110888990A (en) * | 2019-11-22 | 2020-03-17 | 深圳前海微众银行股份有限公司 | Text recommendation method, device, equipment and medium |
| CN110888990B (en) * | 2019-11-22 | 2024-04-12 | 深圳前海微众银行股份有限公司 | Text recommendation method, device, equipment and medium |
| WO2021098648A1 (en) * | 2019-11-22 | 2021-05-27 | 深圳前海微众银行股份有限公司 | Text recommendation method, apparatus and device, and medium |
| CN111190968A (en) * | 2019-12-16 | 2020-05-22 | 北京航天智造科技发展有限公司 | Data preprocessing and content recommendation method based on knowledge graph |
| CN111147375A (en) * | 2019-12-31 | 2020-05-12 | 秒针信息技术有限公司 | Network operation method, device, computer equipment and medium |
| CN111460808A (en) * | 2020-03-23 | 2020-07-28 | 腾讯科技（深圳）有限公司 | Synonymous text recognition and content recommendation method and device and electronic equipment |
| CN111460808B (en) * | 2020-03-23 | 2022-04-26 | 腾讯科技（深圳）有限公司 | Synonymous text recognition and content recommendation method and device and electronic equipment |
| CN113449176A (en) * | 2020-03-24 | 2021-09-28 | 华为技术有限公司 | Recommendation method and device based on knowledge graph |
| CN112163149A (en) * | 2020-09-16 | 2021-01-01 | 北京明略昭辉科技有限公司 | Method and device for recommending messages |
| CN112163149B (en) * | 2020-09-16 | 2024-05-03 | 北京明略昭辉科技有限公司 | Method and device for recommending message |
| CN112214685B (en) * | 2020-09-27 | 2023-03-28 | 电子科技大学 | Knowledge graph-based personalized recommendation method |
| CN112214685A (en) * | 2020-09-27 | 2021-01-12 | 电子科技大学 | Knowledge graph-based personalized recommendation method |
| CN112201348A (en) * | 2020-10-28 | 2021-01-08 | 浙江大学 | Multi-center clinical data set adapting device based on knowledge perception |
| CN112201348B (en) * | 2020-10-28 | 2022-07-26 | 浙江大学 | Knowledge-aware-based multi-center clinical data set adaptation device |
| CN112528010A (en) * | 2020-12-15 | 2021-03-19 | 建信金融科技有限责任公司 | Knowledge recommendation method and device, computer equipment and readable storage medium |
| CN113961823A (en) * | 2021-12-17 | 2022-01-21 | 江西中业智能科技有限公司 | News recommendation method, system, storage medium and equipment |
| CN113961823B (en) * | 2021-12-17 | 2022-03-25 | 江西中业智能科技有限公司 | News recommendation method, system, storage medium and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109165350A (en) | A kind of information recommendation method and system based on deep knowledge perception | |
| US10936906B2 (en) | Training data acquisition method and device, server and storage medium | |
| CN104408153B (en) | A kind of short text Hash learning method based on more granularity topic models | |
| CN103268339B (en) | Named entity recognition method and system in Twitter message | |
| CN110298042A (en) | Based on Bilstm-crf and knowledge mapping video display entity recognition method | |
| CN107578292B (en) | User portrait construction system | |
| CN104573028A (en) | Intelligent question-answer implementing method and system | |
| CN110489523B (en) | Fine-grained emotion analysis method based on online shopping evaluation | |
| CN106021364A (en) | Method and device for establishing picture search correlation prediction model, and picture search method and device | |
| CN110321291A (en) | Test cases intelligent extraction system and method | |
| CN104217008B (en) | Internet personage video interactive mask method and system | |
| CN106202294B (en) | Related news computing method and device based on keyword and topic model fusion | |
| CN110390363A (en) | A kind of Image Description Methods | |
| CN109299245B (en) | Method and device for recalling knowledge points | |
| CN103678271B (en) | A kind of text correction method and subscriber equipment | |
| CN108647225A (en) | A kind of electric business grey black production public sentiment automatic mining method and system | |
| CN108415972A (en) | text emotion processing method | |
| CN110008309A (en) | A kind of short phrase picking method and device | |
| CN111553159B (en) | Question generation method and system | |
| CN113722478B (en) | Multi-dimensional feature fusion similar event calculation method and system and electronic equipment | |
| CN108920451A (en) | Text emotion analysis method based on dynamic threshold and multi-categorizer | |
| US8165987B2 (en) | System and method of machine-aided information extraction rule development | |
| CN117094311B (en) | Method for establishing error correction filter for Chinese grammar error correction | |
| CN112231554B (en) | Search recommended word generation method and device, storage medium and computer equipment | |
| CN113239159A (en) | Cross-modal retrieval method of videos and texts based on relational inference network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190108 |