CN105022760A - News recommendation method and device - Google Patents
News recommendation method and device Download PDFInfo
- Publication number
- CN105022760A CN105022760A CN201410181773.4A CN201410181773A CN105022760A CN 105022760 A CN105022760 A CN 105022760A CN 201410181773 A CN201410181773 A CN 201410181773A CN 105022760 A CN105022760 A CN 105022760A
- Authority
- CN
- China
- Prior art keywords
- interest
- news
- key word
- checks
- weighted value
- 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
Abstract
The invention discloses a news recommendation method. The news recommendation method comprises the following steps: acquiring history records of news which are viewed by all users; determining first view interest weighted values respectively corresponding to the news which are viewed by all users according to a preset rule; extracting key words of all the news; determining second view interest weighted values respectively corresponding to all key words according to the first view interest weighted values; acquiring view interest state transition probabilities according to all the key words and the corresponding second view interest weighted values; determining interest key words of target users according to the view interest state transition probabilities, the key words of the target users and the corresponding second view interest weighted values; and determining recommended news according to the interest key words. Furthermore, the invention also provides a news recommendation device. According to the news recommendation method and device provided by the invention, the matching degree of the recommended news and the interests of the users and the recommendation efficiency can be improved, and the time for searching the news interested by the users is saved.
Description
Technical field
The present invention relates to communication technical field, particularly relate to a kind of news recommend method and device.
Background technology
Along with the development of internet, the custom becoming more and more user of the news that surfs the web, news website initiatively can recommend news to this user.The news of recommending can be up-to-date hot news, also can be recommend news targetedly according to different user.
In prior art, recommend news targetedly to user, based on classical collaborative filtering strategy, its news that may read can be recommended to user.
But in above prior art, because classical collaborative filtering strategy calculates similarity between news based on group of subscribers behavior, then news higher for similarity is recommended the close user of reading habit, but the user behavior of up-to-date news is little, be difficult to the similarity calculating this up-to-date news and other news, so cannot accomplish for different user personalized recommendation latest news.
Summary of the invention
In view of this, the invention provides a kind of news recommend method and device, can check that the different interest of news carry out Personalize News recommendation in conjunction with different user, improve the matching degree of news and the user interest recommended and recommend efficiency, save the time that user finds news interested.
A kind of news recommend method that the embodiment of the present invention provides, comprising: the historical record obtaining the news that each user checks, and determines that news that each user checks corresponding first checks interest weighted value respectively according to presetting rule; Extract the key word of each described news, and check that interest weighted value determines that each described key word corresponding second checks interest weighted value respectively according to described first; According to each described key word and correspondence described second checks that interest state transition probability is checked in the acquisition of interest weighted value; Check interest state transition probability according to described, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of described targeted customer; The news of recommending is determined according to described interest key word.
A kind of news recommendation apparatus that the embodiment of the present invention provides, comprising: acquisition module, determination module, extraction module and acquisition module.Wherein, described acquisition module, for obtaining the historical record of the news that each user checks; Described determination module, for determining that according to presetting rule news that each user checks corresponding first checks interest weighted value respectively; Described extraction module, for extracting the key word of each described news; Described determination module, also for checking that interest weighted value determines that each described key word corresponding second checks interest weighted value respectively according to described first; Described acquisition module, checks interest state transition probability for checking that according to described second of each described key word and correspondence interest weighted value obtains; Described determination module, also for checking interest state transition probability described in basis, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of described targeted customer; Described determination module, also for determining the news of recommending according to described interest key word.
The above-mentioned news recommend method that the embodiment of the present invention provides and device, the historical record of news is checked according to user, determine key word in all news corresponding check interest weighted value, and then calculate that all users check key word check interest state transition probability, interest state transition probability is checked according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determine the interest key word may being interested in check of this targeted customer, and the news recently produced comprising this interest key word is recommended user check, the history news can checked according to multiple user thus infer specific user may interested news, the interest realized based on user recommends news, improve news and recommend the matching degree with user interest, improve the efficiency of recommending news.
For above and other object of the present invention, feature and advantage can be become apparent, preferred embodiment cited below particularly, and coordinate institute's accompanying drawings, be described in detail below.
Accompanying drawing explanation
The applied environment figure of the news recommend method that Fig. 1 provides for the embodiment of the present invention.
Fig. 2 illustrates a kind of server architecture block diagram.
Fig. 3 illustrates a kind of user terminal structured flowchart.
The process flow diagram of the news recommend method that Fig. 4 provides for first embodiment of the invention.
The process flow diagram of the news recommend method that Fig. 5 provides for second embodiment of the invention.
The structural representation of the news recommendation apparatus that Fig. 6 provides for third embodiment of the invention.
Fig. 7 is the storage environment schematic diagram of the device of Fig. 6.
The structural representation of the news recommendation apparatus that Fig. 8 provides for fourth embodiment of the invention.
Embodiment
For further setting forth the present invention for the technological means that realizes predetermined goal of the invention and take and effect, below in conjunction with accompanying drawing and preferred embodiment, to according to the specific embodiment of the present invention, structure, feature and effect thereof, be described in detail as follows.
The news recommend method that the embodiment of the present invention provides can be applicable in the environment shown in Fig. 1, as shown in Figure 1, user terminal 100 and server 200 are arranged in wired or wireless network, and by this wired or wireless network, user terminal 100 and server 200 carry out data interaction.
Particularly, server 200 obtains the historical record of the news that each user checks, and determines that news that each user checks corresponding first checks interest weighted value respectively according to presetting rule; Extract the key word of each this news, and first check that interest weighted value determines that each this key word corresponding second checks interest weighted value respectively according to this; Second check that interest weighted value obtains according to this of respectively this key word and correspondence and check interest state transition probability; Check interest state transition probability according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of this targeted customer; The news of recommending is determined according to this interest key word, and the user terminal 100 sending to targeted customer corresponding the news of this recommendation.The news of recommending for specific user that user terminal 100 reception server 200 sends, and the news of this recommendation is shown to user checks.
Wherein, user terminal 100 can comprise: smart mobile phone, panel computer, E-book reader, MP3 player (Moving Picture Experts Group Audio Layer III, dynamic image expert compression standard audio frequency aspect 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image expert compression standard audio frequency aspect 4) player, pocket computer on knee, vehicle-mounted computer, desk-top computer, Set Top Box, intelligent TV set, wearable device etc.
Fig. 2 shows a kind of structured flowchart of user terminal.As shown in Figure 2, user terminal 100 comprises storer 102, memory controller 104, one or more (only illustrating one in figure) processor 106, Peripheral Interface 108, radio-frequency module 110, locating module 112, photographing module 114, audio-frequency module 116, screen 118 and key-press module 120.These assemblies are by one or more communication bus/signal wire 122 communication mutually.
Be appreciated that the structure shown in Fig. 2 is only signal, user terminal 100 also can comprise than assembly more or less shown in Fig. 2, or has the configuration different from shown in Fig. 2.Each assembly shown in Fig. 2 can adopt hardware, software or its combination to realize.
Storer 102 can be used for storing software program and module, as the news recommend method in the embodiment of the present invention and programmed instruction/module corresponding to device, processor 102 is by running the software program and module that are stored in storer 104, thus perform the application of various function and data processing, namely realize above-mentioned news recommend method.
Storer 102 can comprise high speed random access memory, also can comprise nonvolatile memory, as one or more magnetic storage device, flash memory or other non-volatile solid state memories.In some instances, storer 102 can comprise the storer relative to the long-range setting of processor 106 further, and these remote memories can be connected to user terminal 100 by network.The example of above-mentioned network includes but not limited to internet, intranet, LAN (Local Area Network), mobile radio communication and combination thereof.Processor 106 and other possible assemblies can carry out the access of storer 102 under the control of memory controller 104.
Various softwares in processor 106 run memory 102, instruction are to perform the various function of user terminal 100 and to carry out data processing.
Peripheral Interface 108 is for being coupled to CPU and storer 102 by various external unit.
In certain embodiments, memory controller 104, processor 106 and Peripheral Interface 108 can realize in one single chip.In some other example, they can respectively by independently chip realization.
Radio-frequency module 110, for receiving and sending electromagnetic wave, realizes the mutual conversion of electromagnetic wave and electric signal, thus carries out communication with communication network or other equipment.Radio-frequency module 110 can comprise the various existing circuit component for performing these functions, such as, and antenna, radio-frequency (RF) transceiver, digital signal processor, encrypt/decrypt chip, subscriber identity module (SIM) card, storer etc.Radio-frequency module 110 can with various network as internet, intranet, wireless network carry out communication or carry out communication by wireless network and other equipment.Above-mentioned wireless network can comprise cellular telephone networks, WLAN (wireless local area network) or Metropolitan Area Network (MAN).Above-mentioned wireless network can use various communication standard, agreement and technology, include, but are not limited to global system for mobile communications (Global System for MobileCommunication, GSM), enhancement mode mobile communication technology (Enhanced Data GSMEnvironment, EDGE), Wideband CDMA Technology (wideband code division multipleaccess, W-CDMA), CDMA (Code Division Multiple Access) (Code division access, CDMA), tdma (time division multiple access, TDMA), bluetooth, adopting wireless fidelity technology (Wireless, Fidelity, WiFi) (as IEEE-USA standard IEEE 802.11a, IEEE802.11b, IEEE802.11g and/or IEEE802.11n), the networking telephone (Voice over internetprotocol, VoIP), worldwide interoperability for microwave access (Worldwide Interoperability forMicrowave Access, Wi-Max), other are for mail, the agreement of instant messaging and short message, and any other suitable communications protocol, even can comprise those current agreements be developed not yet.
Locating module 112 is for obtaining the current location of user terminal 100.The example of locating module 112 includes but not limited to Global Positioning System (GPS) (GPS), location technology based on WLAN (wireless local area network) or mobile radio communication.
Photographing module 114 is for taking pictures or video.Photo or the video of shooting can be stored in storer 102, and send by radio-frequency module 110.
Audio-frequency module 116 provides audio interface to user, and it can comprise one or more microphone, one or more loudspeaker and voicefrequency circuit.Voicefrequency circuit receives voice data from Peripheral Interface 108, voice data is converted to telecommunications breath, and telecommunications breath is transferred to loudspeaker.Telecommunications breath is changed the sound wave can heard into people's ear by loudspeaker.Voicefrequency circuit also from microphone receive telecommunications breath, convert electrical signals to voice data, and by data transmission in network telephony to Peripheral Interface 108 to be further processed.Voice data can obtain from storer 102 or by radio-frequency module 110.In addition, voice data also can be stored in storer 102 or by radio-frequency module 110 and send.In some instances, audio-frequency module 116 also can comprise an earphone and broadcast hole, for providing audio interface to earphone or other equipment.
Screen 118 provides an output interface between user terminal 100 and user, and export to user's display video, the content of these video frequency output can comprise word, figure, video and combination in any thereof.Some Output rusults correspond to some user interface object.Understandable, screen 118 can also provide one to export and inputting interface between user terminal 100 and user simultaneously.Particularly, except exporting to user's display video, screen 118 also receives the input of user, and the gesture operation such as click, slip of such as user, so that response is made in the input of user interface object to these users.The technology detecting user's input can be based on resistance-type, condenser type or other touch control detection technology possible arbitrarily.The instantiation of screen 118 display module includes, but are not limited to liquid crystal display or light emitting polymer displays.
Key-press module 120 provides user to carry out the interface inputted to user terminal 100 equally, and user can perform different functions by pressing different buttons to make user terminal 100.
Fig. 3 shows a kind of structured flowchart of server.As shown in Figure 3, server 200 comprises: storer 201, processor 202 and mixed-media network modules mixed-media 203.
Be appreciated that the structure shown in Fig. 3 is only signal, server 200 also can comprise than assembly more or less shown in Fig. 3, or has the configuration different from shown in Fig. 3.Each assembly shown in Fig. 3 can adopt hardware, software or its combination to realize.In addition, the server in the embodiment of the present invention can also comprise the server of multiple concrete difference in functionality.
Storer 201 can be used for storing software program and module, as the news recommend method in the embodiment of the present invention and programmed instruction/module corresponding to device, processor 202 is by running the software program and module that are stored in storer 201, thus perform the application of various function and data processing, namely realize the news recommend method in the embodiment of the present invention.Storer 201 can comprise high speed random access memory, also can comprise nonvolatile memory, as one or more magnetic storage device, flash memory or other non-volatile solid state memories.In some instances, storer 201 can comprise the storer relative to the long-range setting of processor 202 further, and these remote memories can be connected to server 200 by network.Further, above-mentioned software program and module also can comprise: operating system 221 and service module 222.Wherein operating system 221, such as can be LINUX, UNIX, WINDOWS, it can comprise the various component software for management system task (such as memory management, memory device control, power management etc.) and/or driving, and can with various hardware or the mutual communication of component software, thus provide the running environment of other component softwares.Service module 222 operates on the basis of operating system 221, and monitors the request of automatic network by the network service of operating system 221, completes corresponding data processing, and return result to client according to request.That is, service module 222 is for providing services on the Internet to client.
Mixed-media network modules mixed-media 203 is for receiving and sending network signal.Above-mentioned network signal can comprise wireless signal or wire signal.In an example, above-mentioned network signal is cable network signal.Now, mixed-media network modules mixed-media 203 can comprise the elements such as processor, random access memory, converter, crystal oscillator.
First embodiment
Refer to Fig. 4, the process flow diagram of the news recommend method that Fig. 4 provides for first embodiment of the invention.As shown in Figure 4, the news recommend method that the present embodiment provides comprises the following steps:
Step S401, obtains the historical record of the news that each user checks, and determines that news that each user checks corresponding first checks interest weighted value respectively according to presetting rule.
Server 200 obtains the historical record of the news that each user checks, that is, obtain all news that each user checked.And determine that news that each user checks corresponding first checks interest weighted value respectively according to presetting rule, this checks that interest weighted value reflection user checks interest to this news, this news check that interest weighted value is higher, show to this news, this user checks that interest is larger.This presetting rule can be check that the time gap server 200 of this news obtains the distance (namely user checks the time dough softening of the time of this news relative to the present system time of server 200) that user checks the time of the historical record of this news according to user, check the size of performance level, repeating the number of times etc. checked, also can be to reflect to this news, user checks that other parameters that interest is correlated with are determined.
Particularly, each bar news of checking for each user determine respectively each bar news corresponding first check interest weighted value, each bar news and corresponding first checks that the list of the corresponding relation of interest weighted value is as shown in list 1:
List 1
usr：news1，weight1，news2，weight2，news3，weight3…，newsN，weightN
In above formula, usr represents a user, news1, news2, news3 ..., newsN represents each bar news that user usr checked, weight1 be news1 corresponding first check interest weight, weight1 be news1 corresponding first check interest weight, by that analogy, weightN be newsN corresponding first check interest weight, the N in above formula is natural number.
Step S402, extracts the key word of each this news, and first checks that interest weighted value determines that each this key word corresponding second checks interest weighted value respectively according to this.
Server 200 extracts the key word of all news that all users check, all key words extracted are carried out mergings like terms, and according to this first check interest weighted value determine each this key word respectively correspondence second check interest weighted value.
The list of the corresponding relation between the key word of all news that all users check is as shown in list 2:
List 2
news：tag1，tag2，tag3…，tagM
In above formula, news represents all news that all users check, tag represents the key word extracted from each news, key word is included in headline or content, this news can be retrieved according to this key word, a key word is at least comprised in General, key word can be any Chinese, English, numeral, or the mixture of Chinese English digital.Tag1 represents first key word in all news that all users check, tag2 represents second key word in all news that all users check, tag3 represents the 3rd key word in all news that all users check, by that analogy, tagM represents M key word in all news that all users check, M represents the quantity of whole key words of all news that all users check.
Further, the news of checking according to each user corresponding first checks interest weighted value respectively, determine that each this key word corresponding second checks interest weighted value respectively, each key word and first checks that the list of the corresponding relation between interest weighted value is as shown in list 3:
List 3
usr：tag1，weight’1，tag2，weight’2，tag3，weight’3…，tagM，weight’M
In above formula, weight ' 1 represent tag1 corresponding second check interest weighted value, weight ' 2 represent tag2 corresponding second check interest weighted value, by that analogy, weight ' M represent tagM corresponding second check interest weighted value.
It should be noted that, this first checks that interest weighted value is corresponding with each bar news, what represent that this user checks this news checks interest weight, and this second checks that interest weighted value is corresponding with each key word, and what represent that all users check this key word checks interest weight.Particularly, this second checks that interest weighted value can be checked first that interest weighted value is cumulative and obtain, such as, user X checks that first of news A checks that interest weighted value is 0.2, user Y checks that first of news B checks that interest weighted value is 0.3, user Z checks that first of news C checks that interest weighted value is 0.1, and news A, news B, news C contain same key word, so second of this key word check that interest weighted value is 0.2+0.3+0.1=0.6.Certainly, also can be that more complicated algorithm calculates this and second checks interest weighted value, not do concrete restriction herein.
Step S403, according to each key word and correspondence second checks that interest state transition probability is checked in the acquisition of interest weighted value.
According to all users, server 200 checks that all key words of news each self-corresponding second are checked that interest obtains and checked interest state transition probability, this checks that interest state transition probability refers to that all users transfer to the probability of another key word from a key word, reacts the interest state transfer case of all user group's property.
Step S404, checks interest state transition probability according to this, and interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of this targeted customer.
This targeted customer refers to that server 200 is for recommending the user of news to it.Server 200 obtain all key words in whole news that this targeted customer checks and each key word corresponding second check interest weighted value, and check that interest state transition probability, the key word of targeted customer and second of correspondence check interest weighted value according to this, determine the interest key word of this targeted customer, this interest key word refers to the key word that this targeted customer may be interested in check, or can think the key word that may comprise in the news that this targeted customer may be interested in check.
Step S405, determines the news of recommending according to this interest key word.
Server 200 comprises the news of this interest key word according to the interest keyword search of this targeted customer determined, and determine according to presetting rule the news that needs are recommended to this targeted customer, the news of recommending is up-to-date news, namely, be later than the news that preset time produces, this preset time can be arranged arbitrarily by system, such as nearest 24 hours.Particularly, server 200 can in the news being later than preset time generation, inquire about all news comprising this interest key word, and arrange according to news time of origin order, the news of preset quantity nearest for distance present time is defined as the news of recommending, the user terminal all sending to targeted customer corresponding the news of the recommendation determined, is shown to user by this user terminal by the news of the recommendation determined and checks.
In the embodiment of the present invention, the historical record of news is checked according to user, determine key word in all news corresponding check interest weighted value, and then calculate that all users check key word check interest state transition probability, interest state transition probability is checked according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determine the interest key word may being interested in check of this targeted customer, and the news recently produced comprising this interest key word is recommended user check, the history news can checked according to multiple user thus infer specific user may interested news, the interest realized based on user recommends news, improve news and recommend the matching degree with user interest, improve the efficiency of recommending news, save the time that user finds news interested.
Second embodiment.
Refer to Fig. 5, the process flow diagram of the news recommend method that Fig. 5 provides for second embodiment of the invention.As shown in Figure 5, the news recommend method that the present embodiment provides comprises the following steps:
Step S501, obtain the user journal of all users, in this user journal, obtain the historical record of the news that each user checks, and check that the completeness of news determines that each news corresponding first checks interest weighted value respectively according to rise time of each news and user.
User journal can be generated in systems in which, user's operation note is in systems in which recorded in this user journal, wherein also comprise the historical record checking news, server 200 obtains the historical record of the news that each user checks, describes the behavior record of the news that corresponding user checked in this historical record.
Further, according to user, server 200 checks that time of each news and user check that the completeness of news determines that each news corresponding first checks interest weighted value respectively, user checks that the time gap current time of news is nearer, user checks that the completeness of news is higher, first checks that interest weighted value is larger, this news check that interest weighted value is higher, show to this news, this user checks that interest is larger.
Particularly, each bar news of checking for each user determine respectively each bar news corresponding first check interest weighted value, each bar news and corresponding first checks that the list of the corresponding relation of interest weighted value is as shown in list 1:
List 1
usr：news1，weight1，news2，weight2，news3，weight3…，newsN，weightN
In above formula, usr represents a user, news1, news2, news3 ... newsN represents each bar news that user usr checked, Weight1 be news1 corresponding first check interest weight, Weight1 be news1 corresponding first check interest weight, by that analogy, WeightN be newsN corresponding one check interest weight, the N in above formula is natural number.
Step S502, extracts the key word of each news, and first checks that interest weighted value determines that each key word corresponding second checks interest weighted value respectively according to this.
Server 200 extracts the key word of all news that all users check, all key words extracted are carried out mergings like terms, and according to this first check interest weighted value determine each this key word respectively correspondence second check interest weighted value.
The list of the corresponding relation between the key word of all news that all users check is as shown in list 2:
List 2
news：tag1，tag2，tag3…，tagM
In above formula, news represents all news that all users check, tag represents the key word extracted from each news, key word is included in headline or content, this news can be retrieved according to this key word, a key word is at least comprised in General, key word can be any Chinese, English, numeral, or the mixture of Chinese English digital.Tag1 represents first key word in all news that all users check, tag2 represents second key word in all news that all users check, tag3 represents the 3rd key word in all news that all users check, by that analogy, tagM represents M key word in all news that all users check, M represents the quantity of whole key words of all news that all users check.
Further, the news of checking according to each user corresponding first checks interest weighted value respectively, and determine that each this key word corresponding second checks interest weighted value respectively, key word and second checks that the corresponding relation list of interest weighted value is as shown in list 3:
List 3
usr：tag1，weight’1，tag2，weight’2，tag3，weight’3…，tagM，weight’M
In above formula, weight ' 1 represent tag1 corresponding second check interest weighted value, weight ' 2 represent tag2 corresponding second check interest weighted value, by that analogy, weight ' M represent tagM corresponding second check interest weighted value.
It should be noted that, this first checks that interest weighted value is corresponding with each bar news, what represent that this user checks this news checks interest weight, and this second checks that interest weighted value is corresponding with each key word, and what represent that all users check this key word checks interest weight.Particularly, this second checks that interest weighted value can be checked first that interest weighted value is cumulative and obtain, such as, user X checks that first of news A checks that interest weighted value is 0.2, user Y checks that first of news B checks that interest weighted value is 0.3, user Z checks that first of news C checks that interest weighted value is 0.1, and news A, news B, news C contain same key word, so second of this key word check that interest weighted value is 0.2+0.3+0.1=0.6.Certainly, also can be that more complicated algorithm calculates this and second checks interest weighted value, not do concrete restriction herein.
Step S503, obtains and checks interest transition probability in the news that all users check arbitrarily between both keyword.
Server 200 calculates the transition probability in all key words between any both keyword according to association rule mining (Association rule mining), association rule mining is one of most active research method in data mining, can be used for finding that the contact between thing is excavated.
Step S504, using all transition probabilities as element, builds user and checks that the state transition probability matrix of tendency is to obtain checking interest state transition probability.
Particularly, user checks that the state transition probability matrix of tendency can be as follows:
Each row in above state transition probability matrix respectively arranges the transition probability (TP be between both keyword, Transition Probability), represent the probability checking again another key word wherein after user checks one of them key word, that is, comprise one of them key word in the news that user checks and comprise again the probability of another key word.When original state is known, transition probability just can make the prediction of different times.Such as, a
12represent that user transfers to the transition probability (comprise tag1 the news that not only user checks but also comprise the probability of tag2) of tag2 from tag1, a
43represent that user transfers to the transition probability of tag3 from tag4, a
m2represent that user transfers to the transition probability of tag2 from tagM.
Step S505, by being multiplied by this state transition probability matrix, obtaining and checking interest emission probabilities matrix.
Server 200, according to preset transfer step number, by this state transition probability matrix being done this transfer step number power computing, obtaining and checking interest emission probabilities matrix A
(t).That is, t matrix A be multiplied, t represents that user's checks interest degree of divergence, and t is larger, and user checks that interest is loose all the more, and the interest key word of user is more.The concrete numerical value of t can be pre-set by system.This checks interest emission probabilities matrix A
(t)specifically can be expressed as:
Step S506, according to the key word of this targeted customer, interest weighted value checked in the key word of all users and corresponding with it second, and that determines this targeted customer initially checks state vector.
This initially checks that state vector represents that a user checks the second set checking interest weighted value that all key words of news are corresponding, and in this set, second checks with second in previous list 3, the order of interest weighted value checks that the order of interest weighted value is identical.
Server 200 is checked the corresponding relation of interest weighted value based on the key word and second of previous list 3 and puts in order, determine the key word of this targeted customer corresponding second check that interest weighted value and second checks the order of interest weighted value.Particularly, this initially checks that the number of elements in state vector is identical with the number of elements in list 3, if this targeted customer checks in all key words of news certain key word had in list 3, then initially check in state vector and check that interest weighted value is identical with second of this key word same position in list 3, if this targeted customer checks in all key words of news certain key word do not had in list 3, then initially check in state vector and be set to 0 with this key word same position in list 3, namely represent that second of this key word checks that interest weighted value is 0.General, 0 in X is many, because the key word of a user accounts for a little part in the key word of all users, is not that 0 second checks that interest weighted value is less.
In an example, in list of hypotheses 3, comprise tag1 to tag5 totally 5 key words, second of correspondence check that interest weighted value is as shown in the table respectively:
usr：tag1，0.2，tag2，0.3，tag3，0.3，tag4，0.2，tag5，0.4
Targeted customer checks in the key word of news do not have tag3 and tag4, then the state vector X that initially checks of targeted customer is:
X：0.2，0.3，0，0，tag5，0.4
Step S507, initially checks state vector and this interest emission probabilities matrix multiple by this targeted customer, and the interest obtaining this targeted customer checks state vector.
The interest of this targeted customer checks state vector Y=A
(t)x.The interest of this targeted customer check state vector represent through to user check that interest is dispersed after, user checks the second set checking interest weighted value that all key words of news are corresponding, the interest of this targeted customer checks that second in state vector checks that interest weighted value is also the element set with order meaning, initially checks that second in state vector checks that the position of interest weighted value is corresponding with previous list 3 and this.
Owing to passing through and interest emission probabilities matrix multiple, user checks that interest key word has been dispersed, and checks that the quantity of interest key word increases.Further, t value is larger, and represent that user interest is loose all the more, interest key word quantity is more, corresponding, in Y 0 is fewer.
Step S508, check interest weighted value, and the interest of targeted customer checks state vector according to the key word of all users and second of correspondence thereof, obtains the interest key word of targeted customer.
The interest of this targeted customer checks that second in state vector checks that interest weighted value is also the element set with order meaning, check that the position of interest weighted value is corresponding with second in previous list 3, therefore, compared with the interest of this targeted customer being checked state vector checks interest weighted value with second in previous list 3, the interest obtaining this targeted customer check state vector be not 0 position corresponding second check key word that interest weighted value the is corresponding interest key word as targeted customer.
Step S509, by this interest key word respectively corresponding this second check that interest weighted value sorts according to order from big to small.
Server 200 by the interest key word of this targeted customer obtained respectively corresponding this second check that interest weighted value sorts according to order from big to small, second checks that the larger sequence of interest weighted value is more forward.
Step S510, obtains and comprises the news of the described interest key word of sequence preset quantity formerly, will comprise interest key word and meet preset quantity in news database, and the news that news generation time is later than preset time is defined as the news of recommendation.
The news sequence of news database is generally inverted order, that is, the news more produced close to current time sorts more forward.Server 200 obtains the news of this interest key word comprising sequence preset quantity formerly in news database, and interest key word will be comprised meet preset quantity, and the news that news generation time is later than preset time is defined as the news of recommending, such as, more than 3 key words will be comprised, and the news that news generation time is later than on January 1st, 2014 is defined as the news of recommending, and by user terminal that the news of this recommendation sends to targeted customer corresponding.
In the embodiment of the present invention, the historical record of news is checked according to user, determine key word in all news corresponding check interest weighted value, and then calculate that all users check key word check interest state transition probability, interest state transition probability is checked according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determine the interest key word may being interested in check of this targeted customer, and the news recently produced comprising this interest key word is recommended user check, the history news can checked according to multiple user thus infer specific user may interested news, the interest realized based on user recommends news, improve news and recommend the matching degree with user interest, improve the efficiency of recommending news, save the time that user finds news interested.
3rd embodiment
Refer to Fig. 6, the structural representation of the news recommendation apparatus that Fig. 6 provides for third embodiment of the invention.The news recommendation apparatus that the present embodiment provides can run in the server 200 shown in Fig. 1, for realizing the news recommend method in above-described embodiment.As shown in Figure 6, news recommendation apparatus 600 can comprise: acquisition module 601, determination module 602, extraction module 603 and acquisition module 604.
Wherein, acquisition module 601, for obtaining the historical record of the news that each user checks;
Determination module 602, for determining that according to presetting rule news that each user checks corresponding first checks interest weighted value respectively; Also for first checking that interest weighted value determines that each key word corresponding second checks interest weighted value respectively according to this; Also for checking interest state transition probability described in basis, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of described targeted customer; Also for determining the news of recommending according to described interest key word;
Extraction module 603, for extracting the key word of respectively this news;
Obtaining module 604, checking interest state transition probability for checking that according to described second of each described key word and correspondence interest weighted value obtains.
This device can also comprise: sending module (not shown).
Sending module, for the user terminal sending to targeted customer corresponding the news of described recommendation.
Each module can be by software code realization above, and now, above-mentioned each module can be stored in storer 201, as shown in Figure 7.Each module can be realized by hardware such as integrated circuit (IC) chip equally above.
In the present embodiment, each module of news recommendation apparatus 60 realizes the process of respective function, refers to the description of the embodiment of news recommend method shown in earlier figures 4, repeats no more herein.
In the embodiment of the present invention, the historical record of news is checked according to user, determine key word in all news corresponding check interest weighted value, and then calculate that all users check key word check interest state transition probability, interest state transition probability is checked according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determine the interest key word may being interested in check of this targeted customer, and the news recently produced comprising this interest key word is recommended user check, the history news can checked according to multiple user thus infer specific user may interested news, the interest realized based on user recommends news, improve news and recommend the matching degree with user interest, improve the efficiency of recommending news, save the time that user finds news interested.
4th embodiment.
Refer to Fig. 8, the structural representation of the news recommendation apparatus that Fig. 8 provides for fourth embodiment of the invention.The news recommendation apparatus that the present embodiment provides can run in the server 200 shown in Fig. 1, for realizing the news recommend method in above-described embodiment.As shown in Figure 8, the news recommendation apparatus 700 that the present embodiment provides is with the 3rd embodiment difference shown in Fig. 6:
Obtain module 604 also to comprise: the first acquiring unit 6041 and construction unit 6042.
Wherein, the first acquiring unit 6041, for obtaining the transition probability in news that all users check arbitrarily between both keyword;
Construction unit 6042, for using this transition probabilities all as element, build user check that the state transition probability matrix of tendency is to obtain checking interest state transition probability.
Determination module 602 comprises: obtain unit 6021 and the first determining unit 6022.
Wherein, obtaining unit 6021, for by being multiplied by this state transition probability matrix, obtaining the interest emission probabilities matrix checking tendency;
First determining unit 6022, check interest weighted value for the key word and corresponding with it second according to this targeted customer, interest weighted value checked in the key word of all users and corresponding with it second, and that determines this targeted customer initially checks state vector;
First determining unit 6022, also for this targeted customer initially checked state vector and this interest emission probabilities matrix multiple, the interest obtaining this targeted customer checks state vector;
Obtain unit 6021, also for checking interest weighted value according to the key word of all users and second of correspondence thereof, and the interest of targeted customer checks state vector, obtains the interest key word of targeted customer.
Further, acquisition module 601 also for obtaining the user journal of all users, and obtains the historical record of the news that each user checks in this user journal;
Determination module 602, also for checking that according to rise time of each news and user the completeness of news determines that each news corresponding first checks interest weighted value respectively.
Further, determination module 602 also comprises: sequencing unit 6023, second acquisition unit 6024 and the second determining unit 6025.
Sequencing unit 6023, for by this interest key word respectively corresponding this second check that interest weighted value sorts according to order from big to small;
Second acquisition unit 6024, for obtaining the news comprising this interest key word of sequence preset quantity formerly in news database;
Second determining unit 6025, for the news news of sequence preset quantity being formerly defined as recommending.
In the present embodiment, each module of news recommendation apparatus realizes the process of respective function, refers to the description of the embodiment of news recommend method shown in earlier figures 5, Fig. 6, repeats no more herein.
In the embodiment of the present invention, the historical record of news is checked according to user, determine key word in all news corresponding check interest weighted value, and then calculate that all users check key word check interest state transition probability, interest state transition probability is checked according to this, interest weighted value checked in the key word of targeted customer and second of correspondence, determine the interest key word may being interested in check of this targeted customer, and the news recently produced comprising this interest key word is recommended user check, the history news can checked according to multiple user thus infer specific user may interested news, the interest realized based on user recommends news, improve news and recommend the matching degree with user interest, improve the efficiency of recommending news, save the time that user finds news interested.
It should be noted that, in this article, the such as relational terms of first and second grades and so on is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply the relation that there is any this reality between these entities or operation or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thus make to comprise the process of a series of key element, method, article or device and not only comprise those key elements, but also comprise other key elements clearly do not listed, or also comprise by the intrinsic key element of this process, method, article or device.When not more restrictions, the key element limited by statement " comprising ... ", and be not precluded within process, method, article or the device comprising described key element and also there is other identical element.
One of ordinary skill in the art will appreciate that all or part of step realizing above-described embodiment can have been come by hardware, the hardware that also can carry out instruction relevant by program completes, described program can be stored in a kind of computer-readable recording medium, the above-mentioned storage medium mentioned can be ROM (read-only memory), disk or CD etc.
The above, it is only preferred embodiment of the present invention, not any pro forma restriction is done to the present invention, although the present invention discloses as above with preferred embodiment, but and be not used to limit the present invention, any those skilled in the art, do not departing within the scope of technical solution of the present invention, make a little change when the technology contents of above-mentioned announcement can be utilized or be modified to the Equivalent embodiments of equivalent variations, in every case be do not depart from technical solution of the present invention content, according to any simple modification that technical spirit of the present invention is done above embodiment, equivalent variations and modification, all still belong in the scope of technical solution of the present invention.
Claims (14)
1. a news recommend method, is characterized in that, comprising:
Obtain the historical record of the news that each user checks, and determine that news that each user checks corresponding first checks interest weighted value respectively according to presetting rule;
Extract the key word of each described news, and check that interest weighted value determines that each described key word corresponding second checks interest weighted value respectively according to described first;
According to each described key word and correspondence described second checks that interest state transition probability is checked in the acquisition of interest weighted value;
Check interest state transition probability according to described, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of described targeted customer;
The news of recommending is determined according to described interest key word.
2. method according to claim 1, is characterized in that, describedly checks that interest weighted value obtains according to described second of each described key word and correspondence and checks that interest state transition probability comprises:
Obtain the transition probability arbitrarily between both keyword in the news that all users check;
Using all described transition probabilities as element, build user and check that the state transition probability matrix of tendency is to obtain checking interest state transition probability.
3. method according to claim 2, is characterized in that, checks interest state transition probability described in described basis, and interest weighted value checked in the key word of targeted customer and second of correspondence, determines that the interest key word of described targeted customer comprises:
By described state transition probability matrix is multiplied, obtains and check interest emission probabilities matrix.
4. method according to claim 3, is characterized in that, described by being multiplied by described state transition probability matrix, obtains after checking interest emission probabilities matrix and comprises:
Interest weighted value checked in key word and corresponding with it second according to described targeted customer, and interest weighted value checked in the key word of all users and corresponding with it second, and that determines described targeted customer initially checks state vector.
5. method according to claim 4, is characterized in that, described determine described targeted customer initially check state vector after comprise:
Described targeted customer initially checked state vector and described interest emission probabilities matrix multiple, and the interest obtaining described targeted customer checks state vector.
6. method according to claim 5, is characterized in that, described described targeted customer is initially checked state vector and described interest emission probabilities matrix multiple, and the interest obtaining described targeted customer comprises after checking state vector:
Check interest weighted value according to the key word of all users and second of correspondence thereof, and the interest of targeted customer checks state vector, obtain the interest key word of targeted customer.
7. the method according to any one of claim 1 to 6, is characterized in that, the historical record of the news that each user of described acquisition checks, and determines that news that each user checks corresponding first checks that interest weighted value comprises respectively according to presetting rule:
Obtain the user journal of all users, in described user journal, obtain the historical record of the news that each user checks, and check that the completeness of described news determines that each described news corresponding first checks interest weighted value respectively according to rise time of each described news and user.
8. method according to claim 7, is characterized in that, described according to described interest key word determine recommend news comprise:
Described interest key word corresponding described second is checked that interest weighted value sorts according to order from big to small respectively;
Obtain in news database and comprise the news of the described interest key word of sequence preset quantity formerly, will comprise interest key word and meet preset quantity, and the news that news generation time is later than preset time is defined as the news of recommendation.
9. a news recommendation apparatus, is characterized in that, comprising:
Acquisition module, for obtaining the historical record of the news that each user checks;
Determination module, for determining that according to presetting rule news that each user checks corresponding first checks interest weighted value respectively;
Extraction module, for extracting the key word of each described news;
Described determination module, also for checking that interest weighted value determines that each described key word corresponding second checks interest weighted value respectively according to described first;
Obtaining module, checking interest state transition probability for checking that according to described second of each described key word and correspondence interest weighted value obtains;
Described determination module, also for checking interest state transition probability described in basis, interest weighted value checked in the key word of targeted customer and second of correspondence, determines the interest key word of described targeted customer;
Described determination module, also for determining the news of recommending according to described interest key word.
10. device according to claim 9, is characterized in that, described acquisition module also comprises:
First acquiring unit, for obtaining the transition probability in news that all users check arbitrarily between both keyword;
Construction unit, for using all described transition probabilities as element, build user check that the state transition probability matrix of tendency is to obtain checking interest state transition probability.
11. devices according to claim 9 or 10, it is characterized in that, described determination module comprises:
Obtain unit, for by described state transition probability matrix is multiplied, obtains and check interest emission probabilities matrix;
First determining unit, check interest weighted value for the key word and corresponding with it second according to described targeted customer, interest weighted value checked in the key word of all users and corresponding with it second, and that determines described targeted customer initially checks state vector;
Described first determining unit, also for described targeted customer initially checked state vector and described interest emission probabilities matrix multiple, the interest obtaining described targeted customer checks state vector;
Described acquisition unit, also for checking interest weighted value according to the key word of all users and second of correspondence thereof, and the interest of targeted customer checks state vector, obtains the interest key word of targeted customer.
12. devices according to claim 11, is characterized in that,
Described acquisition module, also for obtaining the user journal of all users, and obtains the historical record of the news that each user checks in described user journal.
13. devices according to claim 12, is characterized in that,
Described determination module, also for checking that according to rise time of each described news and user the completeness of described news determines that each described news corresponding first checks interest weighted value respectively.
14. devices according to claim 13, is characterized in that, described determination module also comprises:
Sequencing unit, for corresponding described second checking that interest weighted value sorts according to order from big to small respectively by described interest key word;
Second acquisition unit, for obtaining the news comprising the described interest key word of sequence preset quantity formerly in news database;
Second determining unit, meets preset quantity for comprising interest key word, and the news that news generation time is later than preset time is defined as the news of recommending.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410181773.4A CN105022760B (en) | 2014-04-30 | 2014-04-30 | A kind of news recommended method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410181773.4A CN105022760B (en) | 2014-04-30 | 2014-04-30 | A kind of news recommended method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105022760A true CN105022760A (en) | 2015-11-04 |
| CN105022760B CN105022760B (en) | 2019-06-25 |
Family
ID=54412739
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410181773.4A Active CN105022760B (en) | 2014-04-30 | 2014-04-30 | A kind of news recommended method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105022760B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105279289A (en) * | 2015-12-04 | 2016-01-27 | 中国传媒大学 | Personalized music recommendation sorting method based on exponential decay window |
| CN105701191A (en) * | 2016-01-08 | 2016-06-22 | 腾讯科技（深圳）有限公司 | Push information click rate estimation method and device |
| CN108319628A (en) * | 2017-02-22 | 2018-07-24 | 腾讯科技（深圳）有限公司 | A kind of user interest determines method and device |
| CN108595580A (en) * | 2018-04-17 | 2018-09-28 | 阿里巴巴集团控股有限公司 | News recommends method, apparatus, server and storage medium |
| CN110572521A (en) * | 2019-09-20 | 2019-12-13 | 中国银行股份有限公司 | Voice broadcasting method and device |
| CN111292151A (en) * | 2018-12-07 | 2020-06-16 | 中移互联网有限公司 | Object recommendation method and device, storage medium and object recommendation equipment |
| CN112380449A (en) * | 2020-12-03 | 2021-02-19 | 腾讯科技（深圳）有限公司 | Information recommendation method, model training method and related device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321190A (en) * | 2008-07-04 | 2008-12-10 | 清华大学 | Recommend method and recommend system of heterogeneous network |
| CN102611785A (en) * | 2011-01-20 | 2012-07-25 | 北京邮电大学 | Personalized active news recommending service system and method for mobile phone user |
| CN102982131A (en) * | 2012-11-16 | 2013-03-20 | 杭州东信北邮信息技术有限公司 | Book recommending method based on Markov chain |
| JP2013109427A (en) * | 2011-11-17 | 2013-06-06 | Nippon Telegr & Teleph Corp <Ntt> | Content recommendation device and method and program |
| CN103577579A (en) * | 2013-11-08 | 2014-02-12 | 南方电网科学研究院有限责任公司 | Resource recommendation method and system based on potential demands of users |
| CN103731738A (en) * | 2014-01-23 | 2014-04-16 | 哈尔滨理工大学 | Video recommendation method and device based on user group behavioral analysis |
-
2014
- 2014-04-30 CN CN201410181773.4A patent/CN105022760B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321190A (en) * | 2008-07-04 | 2008-12-10 | 清华大学 | Recommend method and recommend system of heterogeneous network |
| CN102611785A (en) * | 2011-01-20 | 2012-07-25 | 北京邮电大学 | Personalized active news recommending service system and method for mobile phone user |
| JP2013109427A (en) * | 2011-11-17 | 2013-06-06 | Nippon Telegr & Teleph Corp <Ntt> | Content recommendation device and method and program |
| CN102982131A (en) * | 2012-11-16 | 2013-03-20 | 杭州东信北邮信息技术有限公司 | Book recommending method based on Markov chain |
| CN103577579A (en) * | 2013-11-08 | 2014-02-12 | 南方电网科学研究院有限责任公司 | Resource recommendation method and system based on potential demands of users |
| CN103731738A (en) * | 2014-01-23 | 2014-04-16 | 哈尔滨理工大学 | Video recommendation method and device based on user group behavioral analysis |
Non-Patent Citations (1)
| Title |
|---|
| 吴登能: "面向移动互联网的个性化新闻推荐算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105279289A (en) * | 2015-12-04 | 2016-01-27 | 中国传媒大学 | Personalized music recommendation sorting method based on exponential decay window |
| CN105701191A (en) * | 2016-01-08 | 2016-06-22 | 腾讯科技（深圳）有限公司 | Push information click rate estimation method and device |
| CN105701191B (en) * | 2016-01-08 | 2020-12-29 | 腾讯科技（深圳）有限公司 | Pushed information click rate estimation method and device |
| CN108319628A (en) * | 2017-02-22 | 2018-07-24 | 腾讯科技（深圳）有限公司 | A kind of user interest determines method and device |
| CN108595580A (en) * | 2018-04-17 | 2018-09-28 | 阿里巴巴集团控股有限公司 | News recommends method, apparatus, server and storage medium |
| CN108595580B (en) * | 2018-04-17 | 2022-08-09 | 创新先进技术有限公司 | News recommendation method, device, server and storage medium |
| CN111292151A (en) * | 2018-12-07 | 2020-06-16 | 中移互联网有限公司 | Object recommendation method and device, storage medium and object recommendation equipment |
| CN111292151B (en) * | 2018-12-07 | 2023-07-25 | 中移互联网有限公司 | Object recommendation method and device, storage medium and object recommendation equipment |
| CN110572521A (en) * | 2019-09-20 | 2019-12-13 | 中国银行股份有限公司 | Voice broadcasting method and device |
| CN112380449A (en) * | 2020-12-03 | 2021-02-19 | 腾讯科技（深圳）有限公司 | Information recommendation method, model training method and related device |
| CN112380449B (en) * | 2020-12-03 | 2021-11-23 | 腾讯科技（深圳）有限公司 | Information recommendation method, model training method and related device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105022760B (en) | 2019-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10783206B2 (en) | Method and system for recommending text content, and storage medium | |
| CN105022760A (en) | News recommendation method and device | |
| CN108156268B (en) | Method for acquiring device identifier, server and terminal device | |
| CN105094861A (en) | Webpage application program loading method, device and system | |
| US8473512B2 (en) | Dynamic profile slice | |
| CN107341187B (en) | Search processing method, device, equipment and computer storage medium | |
| US9949149B2 (en) | Online and distributed optimization framework for wireless analytics | |
| US20180260855A1 (en) | Method and device for pushing promotion information | |
| CN108763579B (en) | Search content recommendation method and device, terminal device and storage medium | |
| CN104243517A (en) | Content sharing method and device used between different terminals | |
| CN104935744A (en) | Verification code display method, verification code display device and mobile terminal | |
| US20180248879A1 (en) | Method and apparatus for setting access privilege, server and storage medium | |
| US10346737B1 (en) | Distributed multisensor system to record spatially diverse events | |
| CN109995876B (en) | File transmission method, device, system and computer storage medium | |
| CN104717125A (en) | Graphic code storage method and device | |
| CN104462051A (en) | Word segmentation method and device | |
| CN107832148A (en) | Performance optimization method and equipment | |
| CN104917795A (en) | Subscription information alert method and device | |
| CN105095253A (en) | Webpage display method and webpage display device | |
| CN106210094A (en) | Periphery focus road conditions methods of exhibiting, the first client, service end and system | |
| CN105320885A (en) | Method and device for detecting malicious website | |
| EP2958062A1 (en) | Determining multiple users of a network enabled device | |
| CN104426685A (en) | Information management method and apparatus | |
| CN106161575A (en) | User matching method and device | |
| CN108512803A (en) | Remind the method and device of change account bound phone number |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221207 Address after: 1402, Floor 14, Block A, Haina Baichuan Headquarters Building, No. 6, Baoxing Road, Haibin Community, Xin'an Street, Bao'an District, Shenzhen, Guangdong 518100 Patentee after: Shenzhen Yayue Technology Co.,Ltd. Address before: 518057 5-10 building, Fiyta building, Gaoxin Nan Road, Nanshan District hi tech Zone, Shenzhen, Guangdong Patentee before: Shenzhen Tencent Computer System Co.,Ltd. |