CN106599182B - Feature Engineering recommended method and device, video website based on spark streaming real-time streams - Google Patents
Feature Engineering recommended method and device, video website based on spark streaming real-time streams Download PDFInfo
- Publication number
- CN106599182B CN106599182B CN201611147453.2A CN201611147453A CN106599182B CN 106599182 B CN106599182 B CN 106599182B CN 201611147453 A CN201611147453 A CN 201611147453A CN 106599182 B CN106599182 B CN 106599182B
- Authority
- CN
- China
- Prior art keywords
- feature
- flow data
- log
- expansion
- engineering
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/7867—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title and artist information, manually generated time, location and usage information, user ratings
Abstract
The invention discloses a kind of Feature Engineering recommended methods based on spark streaming real-time streams, including, the expansion log and click logs of client are obtained, squeezes into Distributed Message Queue after cleaning；The log stream that expansion log and click logs is subscribed to using spark streaming merges the flow data in two log streams in engineering；Operation flow data generates label, and to identify, expansion is clicked and expansion does not click on flow data；It is that expansion log and click logs construct multidimensional characteristic, while combination foundation temporal characteristics according to foundation characteristic；Flow data with new feature is carried out to off-line training and on-line training respectively to generate recommendation flow data.The present invention provides a kind of feature extraction modes in generally applicable most of fields, solve the problems, such as that the Feature Engineering scope of application is small, and it uses based on online, in a manner of correcting offline, it solves the problems, such as Feature Engineering timeliness, and realizes the validity and accuracy of feature by the feature combined transformation of some column.
Description
Technical field
The present invention relates to video recommendations processing technology fields, real-time based on spark streaming more particularly to one kind
The Feature Engineering recommended method of stream.
Background technique
With comprehensive arrival in 2.0 epoch of internet, how huge using these a large amount of information data being full of in network is
Big and mixed and disorderly data, therefrom digging becomes hot topic according to valuable information out, this is also in data mining as a weight
The machine learning in branch field is wanted to bring the spring of development.In machine learning techniques, few people pay close attention to Feature Engineering
(Feature Engineering), and more remove to take notice of the selection and optimization of model and algorithm, however, being characterized in engineering
The raw material of learning system, the influence to final mask are unquestionable.
Most models can be learnt well by structure good in data, excellent even if not being optimal model
The feature of matter also available good effect.The flexibility of quality features can allow you to use simple model calculation more
Fastly, it is easier to understand, it is easier to safeguard.Have so in short in the industry cycle widespread: data and feature determine machine learning
The upper limit, and model and algorithm only approach this upper limit.
This requires there is process a set of in this way, can effectively carry out feature extraction, and by the primitive character extracted into
A series of processes such as row eigentransformation, feature combination, obtain good feature, Lai Tigao machine learning algorithm model it is accurate
Degree.
Currently, the Feature Engineering construction in many machine learning fields is had existed both at home and abroad, in the recommendation of Meituan order
Feature Engineering, Feature Engineering of Baidu's ad system etc..
The Feature Engineering of Baidu's ad system is mainly in such a way that machine learning algorithm indirect labor extracts, based on big
The experience of amount pick out suitable feature and by a large amount of eigentransformation with combine, realize the high latitude and accuracy of feature,
Rely more on the experience of feature extraction and the united application of many algorithms.
The Feature Engineering construction of the recommender system of Meituan has more the property of electric business, such as is constructed characterized by geographical location
The model etc. that neighbouring businessman recommends.Further according to these features closely coupled with business, using many algorithms to eigentransformation with
Combine the construction to realize Feature Engineering.
Traditional Feature Engineering construction is more dependent on the understanding experience to business, obtains by constantly testing sex exploration
It to the feature for being suitble to itself field, and is that thus there is very strong limitation and lack based on offline feature mostly
Timeliness certainly will cause the operation strategies of Feature Engineering not extensive enough in this way, and difficulty is larger, not be suitable for platform construction and right
Outer popularization.This does not comply with ecology required by nowadays Internet era, hardware and software platform, sharing, timeliness, validity and side
Just property is not able to satisfy video class recommended requirements especially.
Summary of the invention
In view of the technical drawbacks of the prior art, it is an object of the present invention to provide one kind to be based on spark
The Feature Engineering recommended method of streaming real-time streams.
The technical solution adopted to achieve the purpose of the present invention is:
A kind of Feature Engineering recommended method based on spark streaming real-time streams, including,
The expansion log and click logs for obtaining client, squeeze into Distributed Message Queue after cleaning；
The log stream that expansion log and click logs is subscribed to using spark streaming merged for two days in engineering
Flow data in will stream；Operation flow data generates label, and to identify, expansion is clicked and expansion does not click on flow data；
It is that expansion log and click logs construct multidimensional characteristic, while combination foundation temporal characteristics according to foundation characteristic；
It will combine in the output of the flow data after feature hdfs and kafka, carried out in hdfs based on history flow data
GDBT model training simultaneously carries out eigentransformation according to GDBT model and to feature, will be in feature after transformation and hdfs and kafka
Primitive character group, which merges, generates new feature,
Flow data with new feature is carried out to off-line training and on-line training respectively to generate recommendation flow data.
The multidimensional characteristic includes user gradation feature, and video length likes feature, and user characteristics, video card are special
Sign, viewing feature, the basal latency feature include two dimensional features to be limited to the specific time.
CTR prediction model training of the flow data with new feature to carry out LR and FTRL.
Spark streaming receives two log streams and generates the key assignments stream data of unified format, then uses
Union merges two flow data streams, and the flow data stream after merging is operated by reduceByKey and generates label label, occurs
And the video tab being clicked is set to 1 and is otherwise set to 0.
The size of the flow data processing window of spark streaming is set as predetermined time interval, while retaining previous
The displaying flow data of period with the clickstream data delayed to reach to merge.
A kind of Feature Engineering recommendation apparatus based on spark streaming real-time streams, including,
Log collection cleaning module squeezes into distribution after cleaning to obtain the expansion log and click logs of client
Message queue；
Flow data merging module subscribes to the log stream of expansion log and click logs using spark streaming, in work
Merge the flow data in two log streams in journey and operate flow data and generates label to identify expansion click and expansion and not click on
Flow data；
Foundation characteristic constructs module, to be that expansion log and click logs construct multidimensional characteristic according to foundation characteristic,
Online feature construction module is incorporated into the multidimensional characteristic as feature the click time will be unfolded,
Eigentransformation composite module will have been combined in the output of the flow data after feature hdfs and kafka, has been based in hdfs
History flow data carry out GDBT model training simultaneously according to GDBT model and to feature carry out eigentransformation, by feature after transformation with
Primitive character group in hdfs and kafka, which merges, generates new feature, the flow data with new feature is carried out respectively offline
Trained and on-line training is to generate recommendation flow data.
Flow data with new feature is to carry out eigentransformation with the CTR prediction model training of LR and FTRL.
A kind of video website with the Feature Engineering recommendation apparatus.
Compared with prior art, the beneficial effects of the present invention are:
The present invention provides a kind of feature extraction modes in generally applicable most of fields, solve Feature Engineering and are applicable in model
Small problem is enclosed, and is used based on online, in a manner of correcting offline, solves the problems, such as Feature Engineering timeliness, and pass through
The feature combined transformation of some column realizes the validity and accuracy of feature.
Detailed description of the invention
The structure that Fig. 1 show the Feature Engineering recommended method of the invention based on spark streaming real-time streams is shown
It is intended to.
Fig. 2 show flow diagram of the invention.
Specific embodiment
The present invention is described in further detail below in conjunction with the drawings and specific embodiments.It should be appreciated that described herein
Specific embodiment be only used to explain the present invention, be not intended to limit the present invention.
As shown, the present invention is based on the Feature Engineering recommended methods of spark streaming real-time streams, including,
Step 101, the expansion log and click logs for obtaining client, squeeze into Distributed Message Queue after cleaning；
In this step, using the relevant technologies, such as flume, a collection correlation log, the log of collection are buried to client
There are mainly two types of, one is show log, i.e. the exposure log of video；Another kind is click logs, i.e. the feelings that are clicked of video
Condition.And cleaning is carried out to the log of collection, then the log after cleaning is squeezed into MQ, such as kafka.Wherein flow data
Cleaning mainly includes the cleaning of flow data format, unified, the brush amount cheating cleaning of two kinds of journal formats etc..
Step 102, the log stream that expansion log and click logs is subscribed to using spark streaming, is closed in engineering
And the flow data in two log streams；Operation flow data generates label, and to identify, expansion is clicked and expansion does not click on flow data；
In the step, after message queue is squeezed into the log after cleaning, two logs are subscribed to using spark streaming
Stream, in engineering merge two log streams in flow data, according in log user Id and video Id obtain this flow data
Essential characteristic, and label for flow data, label will be used for the CTR prediction model training of subsequent LR and FTRL, then will
As a result it exports.
Specifically, two flow data streams are received using spark streaming and generates the key assignments convection current of unified format
Data show as combination of the unique id of user plus the unique id of video in practice as unique key, the value position 1 of click steam, exhibition
The value for showing stream is 0, is then merged two flow data streams using union, the flow data stream after merging is grasped by reduceByKey
Make to generate label label, occurs and the video tab that was clicked is set to 1 and is otherwise set to 0.
The inconsistent situation of two flow data stream time dimensions in order to prevent, such as the click logs of a video arrive first,
And show that log is less than, the size of flow data processing window is set as 5 minutes, while retaining the displaying fluxion of previous period
According to doing and merge with clickstream data together with the displaying flow data of the period.
It step 103, is that expansion log and click logs construct multidimensional characteristic, while combination foundation time according to foundation characteristic
Feature；The multidimensional characteristic includes user gradation feature, and video length likes feature, and user characteristics, are seen video card feature
Shadow feature, the basal latency feature include two dimensional features to be limited to the specific time.
Firstly, to construct foundation characteristic, the building of foundation characteristic will consider Muhivitamin Formula With Minerals, including,
User gradation feature: 1 dimension according to user's viewing behavior, carries out grade classification to user, is divided into severe and actively uses
Family, slight any active ues, ordinary user, less any active ues, inactive user, 5 major class are marked with 4,3,2,1,0 respectively
Know, note: 4 be any active ues 0 be inactive user.
1 dimension, short-sighted frequency hobby: the short number of videos of viewing in nearest January.
Long video hobby: 1 dimension watches long video quantity in nearest January.
300 dimensions, user's topic model: user characteristics, this feature pass through word2vec by watched video in user 10 days
Obtained feature is composed.
Video card feature: 300 dimensions are composed of the video that card encapsulates by the feature that word2vec is obtained.
168 dimensions, user's viewing feature: user 7*24 hour, viewing situation, i.e. interior viewing video counts hourly weekly
Amount.
Different fine-grained dimensions are processed with respect to flow data, and subsequent processing is effectively ensured.It is special on above-mentioned basis simultaneously
Further include bidimensional basal latency feature except sign, You Yizhou when that minute (hour* in (dayOfWeek) and one day
It 60+minute) forms, be used to record specific expansion or click the time, construct 773 dimensional features in this way.
Step 104, the flow data after feature will have been combined to export respectively and carry out eigentransformation to feature, it then will transformation
Feature merges with primitive character group afterwards generates new feature,
It in the step, flow data after having combined feature while exporting in hdfs and kafka, wherein kafka is used for
Online model training, the continuous flow data of real-time reception simultaneously carry out model training, flow data format be similar to " label,
Features ", hdfs are used for offline model training, to receive data and utilize the history in previous predetermined amount of time
Data carry out model training, do not have real-time, flow data format be similar to " ukey t pdna t label t
Features ", wherein features is basic feature, using the format of key-value pair, be similar to " 1:0.1 ... 773:123) ".
According to the history flow data being output in hdfs, the training of the history flow data progress GDBT model such as in two weeks,
And select the good training pattern of an effect as a result, then with this trained model to online and offline two flow datas into
Row eigentransformation generates 1280 new dimensional features, and it is grouped together with original 773 dimensional feature and is ultimately formed after transformation
2013 dimensional features,
Step 105, off-line training and on-line training will be carried out to generate recommendation flow data with the flow data of new feature.
In the step, then using GDBT model to the flow data after having combined feature, i.e. fluxion in hdfs and kafka
According to progress eigentransformation with the CTR prediction model of offline LR and online FTRL.
Traditional characteristic engineering mostly uses greatly offline means, is based on history flow data construction feature, this latent structure mode
The model built, which does not account for flow data, has the characteristics that real-time variation, and modelling effect is to be improved.And this base
It again can not be by way of reducing sample training time interval come real in the case where big flow data amount in the training of full dose feature
Existing timeliness feature.The present invention is based on spark streaming real-time streaming processing modes, provide a kind of on-line/off-line
(online/offline) mode combined carries out feature extraction, combination, transformation, uses this side online based on increment
Offline correct of the problem of formula very good solution training sample timeliness, additional offline enable feature samples to reach
Very high effect.
GDBT model training is carried out with data flow data first one or two weeks ago, then with trained model to off-line data
Flow data and online data flow data carry out eigentransformation, then transformed feature, and it is pre- that offline feature is output to LR progress CTR
It surveys, online feature is output to FTRL and carries out CTR prediction.By the training of GDBT model, historical data is realized and in line number
According to interaction, as spark streaming processing log stream beats label (label whether clicked) to every data flow data
When, it is possible that click logs and displaying log have the time difference to lead to the situation for judging whether to click inaccuracy, correct
Data can not embody in on-time model again, but the data flow data corrected is output to GDBT, so that it may use off-line data
Flow data is corrected, and then can just train relatively accurate model with such.It is fed back eventually by the form of changing features
To on-time model.
Meanwhile the invention also discloses the Feature Engineering recommendation apparatus based on spark streaming real-time streams, including,
Log collection cleaning module squeezes into distribution after cleaning to obtain the expansion log and click logs of client
Message queue；
Flow data merging module subscribes to the log stream of expansion log and click logs using spark streaming, in work
Merge the flow data in two log streams in journey and operate flow data and generates label to identify expansion click and expansion and not click on
Flow data；
Foundation characteristic constructs module, to be that expansion log and click logs construct multidimensional characteristic according to foundation characteristic,
Online feature construction module is incorporated into the multidimensional characteristic, combination as feature the click time will be unfolded
Flow data after complete feature exports in hdfs and kafka simultaneously, carries out GDBT mould according to the history flow data being output in hdfs
The training of type, then using GDBT model to the flow data after having combined feature carry out eigentransformation with offline LR training and
The FTRL training of line.
Eigentransformation composite module carries out eigentransformation, feature after transformation to the multidimensional characteristic after stream data combination
Merge with primitive character group and generate new feature, off-line training module and on-line training module are based respectively on new feature and carry out
Line prediction and offline prediction are to generate recommendation flow data.
The present invention provides a kind of feature extraction modes in generally applicable most of fields, solve Feature Engineering and are applicable in model
Small problem is enclosed, and is used based on online, in a manner of correcting offline, solves the problems, such as Feature Engineering timeliness, and pass through
The feature combined transformation of some column realizes the validity and accuracy of feature.
The above is only a preferred embodiment of the present invention, it is noted that for the common skill of the art
For art personnel, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
Also it should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of Feature Engineering recommended method based on spark streaming real-time streams, which is characterized in that including,
The expansion log and click logs for obtaining client, squeeze into Distributed Message Queue after cleaning；
The log stream that expansion log and click logs is subscribed to using spark streaming merges two log streams in engineering
In flow data；Operation flow data generates label, and to identify, expansion is clicked and expansion does not click on flow data；
It is that expansion log and click logs construct multidimensional characteristic, while combination foundation temporal characteristics according to foundation characteristic；
It will combine in the output of the flow data after feature hdfs and kafka, GDBT mould carried out based on history flow data in hdfs
Type training simultaneously carries out eigentransformation to feature according to GDBT model, by the primitive character in feature after transformation and hdfs and kafka
Group, which merges, generates new feature,
Flow data with new feature is carried out to off-line training and on-line training respectively to generate recommendation flow data.
2. the Feature Engineering recommended method as described in claim 1 based on spark streaming real-time streams, feature exist
In the multidimensional characteristic includes user gradation feature, video length hobby feature, user characteristics, video card feature and sight
Shadow feature, the basal latency feature include two dimensional features to be limited to the specific time.
3. the Feature Engineering recommended method as described in claim 1 based on spark streaming real-time streams, feature exist
In CTR prediction model training of the flow data with new feature to carry out LR and FTRL.
4. the Feature Engineering recommended method as described in claim 1 based on spark streaming real-time streams, feature exist
In spark streaming receives two log streams and generates the key assignments stream data of unified format, then will using union
Two flow datas merge, and the flow data after merging is operated by reduceByKey and generates label label, occur and are clicked
Video tab be set to 1 and be otherwise set to 0.
5. the Feature Engineering recommended method as described in claim 1 based on spark streaming real-time streams, feature exist
In the size of the flow data processing window of spark streaming is set as predetermined time interval, while retaining the previous period
Show the flow data of log to merge with the clickstream data delayed to reach.
6. a kind of Feature Engineering recommendation apparatus based on spark streaming real-time streams, which is characterized in that including,
Log collection cleaning module squeezes into distributed message after cleaning to obtain the expansion log and click logs of client
Queue；
Flow data merging module subscribes to the log stream of expansion log and click logs using spark streaming, in engineering
Merge the flow data in two log streams and operate flow data and generates label to identify expansion click and expansion and not click on fluxion
According to；
Foundation characteristic constructs module, to be that expansion log and click logs construct multidimensional characteristic according to foundation characteristic,
Online feature construction module is incorporated into the multidimensional characteristic as feature the click time will be unfolded,
Eigentransformation composite module will combine in the output of the flow data after feature hdfs and kafka, history be based in hdfs
Flow data carries out GDBT model training and simultaneously carries out eigentransformation to feature according to GDBT model, by feature after transformation and hdfs and
Primitive character group in kafka, which merges, generates new feature, the flow data with new feature is carried out respectively off-line training and
On-line training is to generate recommendation flow data.
7. the Feature Engineering recommendation apparatus as claimed in claim 6 based on spark streaming real-time streams, feature exist
In the flow data with new feature is to carry out eigentransformation with the CTR prediction model training of LR and FTRL.
8. a kind of video website with Feature Engineering recommendation apparatus as claimed in claims 6 or 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611147453.2A CN106599182B (en) | 2016-12-13 | 2016-12-13 | Feature Engineering recommended method and device, video website based on spark streaming real-time streams |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611147453.2A CN106599182B (en) | 2016-12-13 | 2016-12-13 | Feature Engineering recommended method and device, video website based on spark streaming real-time streams |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106599182A CN106599182A (en) | 2017-04-26 |
| CN106599182B true CN106599182B (en) | 2019-10-11 |
Family
ID=58801022
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611147453.2A Active CN106599182B (en) | 2016-12-13 | 2016-12-13 | Feature Engineering recommended method and device, video website based on spark streaming real-time streams |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106599182B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704594B (en) * | 2017-10-13 | 2021-02-09 | 东南大学 | Real-time processing method for log data of power system based on spark streaming |
| CN110704551B (en) * | 2018-06-21 | 2023-02-17 | 中兴通讯股份有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN110690984A (en) * | 2018-07-05 | 2020-01-14 | 上海宝信软件股份有限公司 | Spark-based big data weblog acquisition, analysis and early warning method and system |
| CN109635948A (en) * | 2018-12-19 | 2019-04-16 | 北京达佳互联信息技术有限公司 | On-line training method, apparatus, system and computer readable storage medium |
| CN110009484A (en) * | 2019-03-12 | 2019-07-12 | 平安普惠企业管理有限公司 | Business data processing method, equipment, server and computer readable storage medium |
| CN111210156B (en) * | 2020-01-13 | 2022-04-01 | 拉扎斯网络科技（上海）有限公司 | Real-time stream data processing method and device based on stream window |
| CN114822855B (en) * | 2022-06-28 | 2022-09-20 | 北京智精灵科技有限公司 | Cognitive training task pushing method, system and construction method based on FTRL model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9032435B2 (en) * | 2011-03-29 | 2015-05-12 | Hulu, LLC | Ad selection and next video recommendation in a video streaming system exclusive of user identity-based parameter |
| CN105654125A (en) * | 2015-12-29 | 2016-06-08 | 山东大学 | Method for calculating video similarity |
-
2016
- 2016-12-13 CN CN201611147453.2A patent/CN106599182B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9032435B2 (en) * | 2011-03-29 | 2015-05-12 | Hulu, LLC | Ad selection and next video recommendation in a video streaming system exclusive of user identity-based parameter |
| CN105654125A (en) * | 2015-12-29 | 2016-06-08 | 山东大学 | Method for calculating video similarity |
Non-Patent Citations (2)
| Title |
|---|
| 基于HBase的实时消息推荐系统;郭若男等;《电信工程技术与标准化》;20141130;第2014卷(第11期);第78-81页 * |
| 基于SPARK平台的实时流计算推荐系统的研究与实现;张贤德;《中国优秀硕士学位论文全文数据库信息科技辑》;20161115;第2016卷(第11期);第I138-498页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106599182A (en) | 2017-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106599182B (en) | Feature Engineering recommended method and device, video website based on spark streaming real-time streams | |
| CN103189856B (en) | The method and apparatus determining media impression | |
| CN107852252B (en) | Fingerprint matching is recognized by automated content to change to detect channel | |
| Rydin Gorjão et al. | Open database analysis of scaling and spatio-temporal properties of power grid frequencies | |
| CN108446964B (en) | User recommendation method based on mobile traffic DPI data | |
| CN106851349A (en) | Based on magnanimity across the live recommendation method for shielding viewing behavior data | |
| CN103870454A (en) | Method and method for recommending data | |
| Holtkamp et al. | Regional patterns of food safety in China: What can we learn from media data? | |
| CN103136280A (en) | System and method of automatically searching hot point video | |
| CN104811810A (en) | Real-time regional audience rating and audience share statistical system based on intelligent television and method thereof | |
| Ma et al. | Review of family-level short-term load forecasting and its application in household energy management system | |
| CN106303591B (en) | A kind of video recommendation method and device | |
| Chen | Do inland provinces benefit from coastal foreign direct investment in China? | |
| CN104866490B (en) | A kind of video intelligent recommended method and its system | |
| Zhao et al. | Research on sales forecast based on prophet-SARIMA combination model | |
| CN104657477A (en) | Social contact recommendation method and device | |
| CN116432998A (en) | Big data internal feedback method and system for water quality anomaly monitoring | |
| Guo et al. | Research on personalized courseware recommendation system of rural distance learning based on combination recommendation technology | |
| Fallon et al. | Understanding Climate Risk in Future Energy Systems: An Energy–Climate Data Hackathon | |
| Bogdanović et al. | One solution to the system architecture for the acquisition of data obtained by monitoring computer users | |
| CN106454400B (en) | A kind of timing subject matter preferences prediction technique of network-oriented TV user | |
| Niu | Smart supervision based on a new generation of information technology integration and its application | |
| Backstrom et al. | Optimizing web traffic via the media scheduling problem | |
| Liu et al. | Research and implementation of POI recommendation system integrating temporal feature | |
| Haider et al. | Big Data in Internet of Things: Architecture and Open Research Challenges |
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 |