CN104504032B - The method and apparatus for being inclined to the service of offer based on user feeling - Google Patents
The method and apparatus for being inclined to the service of offer based on user feeling Download PDFInfo
- Publication number
- CN104504032B CN104504032B CN201410773678.3A CN201410773678A CN104504032B CN 104504032 B CN104504032 B CN 104504032B CN 201410773678 A CN201410773678 A CN 201410773678A CN 104504032 B CN104504032 B CN 104504032B
- Authority
- CN
- China
- Prior art keywords
- text
- assessed
- user
- time
- emotion
- 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
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
Embodiments herein discloses a kind of method for being inclined to the service of offer based on user feeling characterized by comprising determines text to be assessed, wherein the text to be assessed is at least text that a user read；According to the model that a user feeling develops at any time, the probability distribution that the text to be assessed enables the user generate each Sentiment orientation is determined；It enables the user generate the probability distribution of each Sentiment orientation according to the text to be assessed, provides service for the user.Disclosed herein as well is a kind of equipment for being inclined to the service of offer based on user feeling.Using the method and apparatus described herein for being inclined to the service of offer based on user feeling, user's feelings can be inclined to and carry out depth analysis, and targetedly service is provided accordingly, compensate for the defect of the prior art.
Description
Technical field
This application involves data mining technology field more particularly to a kind of methods for being inclined to the service of offer based on user feeling
And equipment.
Background technique
In recent years, with the continuous development of internet the relevant technologies, various content push are increasingly becoming application provider's increasing
Add the main means of user's frequency of use.Currently there are much content push modes based on user behavior.For example, clear according to user
The commodity look at push similar merchandise promotion information to user；Clothes according to the current location of user, near user's push
Information of being engaged in etc..
Currently, many portal websites start the mood for allowing user to share them about some news.By analyzing user
To the Sentiment orientation of some news, it is known that user sees that news can be glad, sees that news can be angry, what news seen
Can be dejected etc..
Which type of Sentiment orientation user generates after reading text, largely reflects the point of interest of user
Place.However, existing content push mode has certain limitation, the Sentiment orientation of user drilling at any time can not be known
Become, it is even more impossible to the push of content is carried out based on this.
Summary of the invention
The purpose of the application is: providing a kind of method and apparatus for being inclined to the service of offer based on user feeling.
According to the one aspect of at least one embodiment of the application, provide a kind of based on user feeling tendency offer service
Method, comprising:
Determine text to be assessed, wherein the text to be assessed is at least text that a user read；
According to the model that a user feeling develops at any time, determine that the text to be assessed enables the user generate each
The probability distribution of Sentiment orientation；
It enables the user generate the probability distribution of each Sentiment orientation according to the text to be assessed, is mentioned for the user
For service.
According to the other side of at least one embodiment of the application, provide a kind of based on user feeling tendency offer service
Equipment, comprising:
One assessment text determining device, for determining text to be assessed, wherein the text to be assessed is that a user reads
At least text crossed；
One emotion probability distribution determining device, the model for being developed at any time according to a user feeling, determine it is described to
Assess the probability distribution that text enables the user generate each Sentiment orientation；
One service providing apparatus, for enabling the user generate the general of each Sentiment orientation according to the text to be assessed
Rate distribution, provides service for the user.
Using the method and apparatus described herein for being inclined to the service of offer based on user feeling, user's feelings can be inclined to
Depth analysis is carried out, and targetedly service is provided accordingly, compensates for the defect of the prior art.
Detailed description of the invention
Fig. 1 is the method flow schematic diagram for establishing emotion model that one embodiment of the application provides；
Fig. 2 is the interactive interface schematic diagram of user feedback Sentiment orientation in one embodiment of the application；
Fig. 3 a is the method flow schematic diagram for establishing emotion model that another embodiment of the application provides；
Fig. 3 b is the method flow schematic diagram for establishing emotion model that another embodiment of the application provides；
Fig. 4 is the method flow schematic diagram for the prediction user feeling tendency that the application one embodiment provides；
Fig. 5 is the method flow schematic diagram for the monitoring public sentiment that the application one embodiment provides；
Fig. 6 is the method flow schematic diagram that the service of offer is inclined to based on user feeling that the application one embodiment provides；
Fig. 7 is the apparatus structure schematic diagram for establishing emotion model that the application one embodiment provides；
Fig. 8 is the apparatus structure schematic diagram for establishing emotion model that another embodiment of the application provides；
Fig. 9 is the apparatus structure schematic diagram for establishing emotion model that another embodiment of the application provides；
Figure 10 is the apparatus structure schematic diagram for establishing emotion model that another embodiment of the application provides；
Figure 11 is 748 structure of the first output sub-module 744 or the second output sub-module that the application one embodiment provides
Schematic diagram；
Figure 12 is the device structure schematic diagram for the prediction user feeling tendency that the application one embodiment provides；
Figure 13 is the device structure schematic diagram for the prediction user feeling tendency that another embodiment of the application provides；
Figure 14 is the device structure schematic diagram for the prediction user feeling tendency that another embodiment of the application provides；
Figure 15 is the device structure schematic diagram for the prediction user feeling tendency that another embodiment of the application provides；
Figure 16 is the device structure schematic diagram for the monitoring public sentiment that the application one embodiment provides；
Figure 17 is the device structure schematic diagram for the monitoring public sentiment that another embodiment of the application provides；
Figure 18 is the device structure schematic diagram for the monitoring public sentiment that another embodiment of the application provides；
Figure 19 is the device structure schematic diagram for the monitoring public sentiment that another embodiment of the application provides；
Figure 20 is the device structure schematic diagram for the monitoring public sentiment that another embodiment of the application provides；
Figure 21 is the device structure schematic diagram that the service of offer is inclined to based on user feeling that the application one embodiment provides；
Figure 22 is the device structure signal that the service of offer is inclined to based on user feeling that another embodiment of the application provides
Figure；
Figure 23 is the device structure signal that the service of offer is inclined to based on user feeling that another embodiment of the application provides
Figure；
Figure 24 is the device structure signal that the service of offer is inclined to based on user feeling that another embodiment of the application provides
Figure；
Figure 25 is the device structure signal that the service of offer is inclined to based on user feeling that another embodiment of the application provides
Figure.
Specific embodiment
With reference to the accompanying drawings and examples, the specific embodiment of the application is described in further detail.Implement below
Example is not limited to scope of the present application for illustrating the application.
It will be appreciated by those skilled in the art that the size of the serial number of following each steps is not intended in embodiments herein
Execution sequence it is successive, the execution of each step sequence should be determined by its function and internal logic, and implement without coping with the application
The implementation process of example constitutes any restriction.
In addition, the terms such as " first ", " second " in the application are only used for difference different step, equipment or module etc., both
Any particular technology meaning is not represented, does not indicate the inevitable logical order between them yet.
Fig. 1 is the method that emotion model is established described in the application one embodiment, referring to Fig. 1, which comprises
S100: an at least training text is determined；
S120: pre-processing an at least training text, determines that the text vector sum of each training text is every
The emotion vector of one training text；
S140: according to the emotion vector of the text vector of each training text, each training text with
And the issuing time of each training text, determine the model that user feeling develops at any time.
Optionally, in the alternative embodiment of the application, an at least training text is determined in above-mentioned S100, can wrap
It includes: the content issued in a period of time is obtained from internet, as training text.It is of course also possible to be with other modes
Training text is obtained, for example, user is directed into a few text as training text manually, embodiments herein does not limit this
It is fixed.Above content may include: news, social networks speech (blog or microblogging etc.) or the model of network forum etc.
Deng.
Obtain training text after, so that it may pre-processed, determine each training text text vector sum each
The emotion vector (S120) of training text.
Optionally, above-mentioned text vector can be effective word of each training text.For example, each training text
Text vector can indicate are as follows:
D={ w1, w2, w3, w1, w4, w5, w5, w2 ... }.
Optionally, above-mentioned emotion vector can read at least one generated after each training text for user in advance
Sentiment orientation.For example, the emotion vector of each training text can indicate are as follows:
Wherein, n1, n2, n3, n4 ... can indicate to generate the quantity of the Sentiment orientation, illustratively, above-mentioned a certain kind feelings
The quantity for feeling tendency can be the statistical magnitude for generating the Sentiment orientation, alternatively, can also be the normalizing that the Sentiment orientation generates
Change quantity, for example, the statistical magnitude that the normalization quantity can be Sentiment orientation generation accounts for all Sentiment orientation statistical magnitudes
Ratio.
For text vector, it is generally the case that in a text, get rid of some meaningless words (for example,
" " word of " " etc) after, remaining is exactly effective word.
And for emotion vector, user can pass through interactive interface as shown in Figure 2 after reading training text
Selection can most represent the option of oneself Sentiment orientation, therefore can be counted according to the data of user feedback, obtain each
The emotion vector of a training text.Or after user's reading training text, commenting for oneself can be delivered by forms such as texts
By classifying to the comment of user, so that it may obtain the emotion vector of each training text.
In one alternative embodiment of the application, in each training of the text vector sum that each training text has been determined
After the emotion vector of text, by each of text vector of each training text effectively word and each training text
Training text emotion vector and issuing time binding, obtain a tuple of training text:If training
Text d has Nd word, then the tuple of training text dAlso there are Nd.In all tuples of a training text
In, emotion vectorIt is all identical with issuing time t.Above-mentioned issuing time can be specific to " hour " or " day ", certainly
Can also be larger with granularity, for example, embodiments herein is not made this specifically specific to " moon " or specific to " year " etc.
It limits.
In another alternative embodiment of the application, training text can be divided according to issuing time, be belonged to
In the training text in section at the same time, issuing time may be considered identical.For belonging in section at the same time
Training text individually establish submodel, then user feeling described herein can include multiple to the model of time-evolution
Submodel in different time periods.In each submodel, due to the issuing time of training text be it is identical, obtain
Tuple can only include effective word and emotion vector, it may be assumed that
In embodiments herein, it is assumed that text is made of all kinds of themes, and " theme " expression one herein is general
It reads, on one side, for image, theme may be considered a bucket, and the inside has filled the higher tuple of probability of occurrence, these tuples
There is very strong correlation with this theme.Pass through " theme " this intermediary, so that it may connect text and tuple.
Therefore, in embodiments herein, some text generate a certain tuple probability can by " text with
Certain probability selection some theme, this theme produce a certain tuple with certain probability " such a process obtains
's.That is:
P (tuple | text)=ΣThemeP (tuple | theme) × P (theme | text)
Each training text is counted, determines the number that some tuple occurs in some training text, so that it may
To obtain the P (tuple | training text) based on training text.Then by an algorithm, obtain P (tuple | theme) and P (theme |
Training text).The P (tuple | theme) and P (theme | training text) obtained according to training, so that it may determine any text generation
The approximate Posterior distrbutionp P (tuple | text) of any tuple.
In one alternate embodiment, as shown in Figure 3a, according to the text of each training text in above-mentioned S140
Vector, the emotion vector of each training text and the issuing time of each training text, determine user feeling with
The model of time-evolution may include:
S141: by the emotion of each of text vector of i-th of training text effectively word and i-th of training text
The binding of the issuing time of vector and i-th of training text, obtains multiple tuples of i-th of training text；
S142: counting i-th of training text, determines occur the probability P of tuple r in i-th of training text
(tuple r | training text i)；
S143: according to the P (tuple r | training text i) determines i-th of training text selection by an algorithm
Theme k probability P (theme k | training text i) and the theme k generate tuple r probability P (tuple r | theme k)；
S144: for according to the P (theme k | training text i) and the P (tuple r | theme k) determines any text
Generate the approximate Posterior distrbutionp P (tuple | text) of any tuple.
In another alternative embodiment, as shown in Figure 3b, according to the text of each training text in above-mentioned S140
Word vector, the emotion vector of each training text and the issuing time of each training text, determine user feeling
The model developed at any time may include:
S145: by the feelings of each of text vector of each training text effectively word and each training text
Feel vector binding, obtains multiple tuples of each training text；
S146: to an at least period, the training text of publication is counted, and determines publication in at least period
I-th of training text in occur tuple r probability P (tuple r | training text i)；
S147: according to the P (tuple r | training text i) determines i-th of training text selection by an algorithm
Theme k probability P (theme k | training text i) and the theme k generate tuple r probability P (tuple r | theme k)；
S148: according to the P (theme k | training text i) and the P (tuple r | theme k) determines an at least period
Interior any text generates the approximate Posterior distrbutionp P (tuple | text) of any tuple.
Optionally, above-mentioned algorithm can be the algorithm based on sampling, or be also possible to variation EM (Expectation-
Maximization, expectation maximization) algorithm.Algorithm based on sampling passes through the sample for collecting Posterior distrbutionp, with point of sample
Cloth acquires the approximation of Posterior distrbutionp, and the common algorithm based on sampling is for example based on gibbs sampler (Gibbs Sampling)
Algorithm.And variation EM algorithm is the distribution for first assuming parametrization of the family on concealed structure, then is changed by variation thought
Generation, which updates, to be found and the immediate distribution of Posterior distrbutionp.Below by taking the algorithm based on gibbs sampler as an example, the application reality is introduced
Apply two kinds of processes that model is determined in example.
(1) to assume a training text collection D, there is m training text d1, d2, d3 ... dm, issuing time is respectively as follows: t1,
T2, t3 ... tm, n theme Z1, Z2, Z3 ... Zn.
Each of training text collection training text is all converted to the form of tuple:
When initial: with impartial probabilityOr a master is assigned for each tuple of each training text at random
Topic, illustratively, as shown in table 1:
Table 1
Situation is assigned according to initial theme, is the tuple in i-th of training text diTheme probability is calculated,
That is: the theme assignment for excluding the tuple, according to other all tuples (other tuples including i-th training text and other
All tuples of training text) theme assignment, estimate text di in tupleIt is endowed the probability of theme Zk
Obtain the tupleIt is again the tuple according to these probability after belonging to the probability of each themeAssign a theme.
Then the theme that next tuple is updated with same method, until finding the theme distribution under each training text
Distribution P (tuple | theme) convergence of P (theme | training text) and the tuple under each theme, algorithm stop.Illustratively, it calculates
Method stop after, theme assign the case where it is as shown in table 2:
Table 2
Then the case where can assigning according to newest theme, the statistics of quantity is carried out, determines that any text generates and appoints
The approximate Posterior distrbutionp P (tuple | text) of one tuple, for example, can count:
There is the probability P (theme | text) of any theme in any text；
Any theme using any effective word probability P (effective word | theme)；
Any theme generates the probability P (Sentiment orientation | theme) of any Sentiment orientation；And
Any theme results from the probability P (time | theme) of any one time.
Above-mentioned probability distribution, so that it may be considered the model that user feeling described in the embodiment of the present application develops at any time
Output.
(2) assume multiple training text subset Dst1, Dt2, Dt3……Dtm, each training text subset includes at least one trained
Text, each training text subset issuing time are respectively as follows: t1, t2, t3 ... tm, have n1 master in training text subset D 1
It inscribes, has n2 theme in training text subset D 2, there is n3 theme ... in training text subset D 3.Wherein, each training text
The corresponding theme of this subset can be different, and the quantity of theme is also possible to different.
Due to the process that model is established according to each training text subset be it is identical, below just just for one
A training text subset describes in detail.
Each of some training text subset training text is all converted to the form of tuple:
It is similar with a upper process, it is at random that each tuple of each training text assigns a theme when initial；
Situation is assigned according to initial theme, is a certain tuple in some training text diCalculate theme probability；It obtains
The tupleIt is again the tuple according to this probability distribution after belonging to the probability of each themeAssign a master
Topic.Then the theme that next tuple is updated with same method, until finding that the theme distribution P under each training text is (main
Topic | text) and tuple under each theme distribution P (tuple | theme) convergence, algorithm stopping.
Then the case where can assigning according to newest theme, the statistics of quantity is carried out, determined in an at least period
Any text generates the approximate Posterior distrbutionp P (tuple | text) of any tuple, for example, can count:
There is the probability P (theme | text) of any theme in any text；
Any theme using any effective word probability P (effective word | theme)；And
Any theme generates the probability P (Sentiment orientation | theme) of any Sentiment orientation.
Since training text subset was divided according to the time, it is built when for each training text subset
After erecting submodel, so that it may obtain the model of user feeling Temporal Evolution.
Therefore, in one alternative embodiment of the application, determine that any text generates any member in above-mentioned S144 or S148
The approximate Posterior distrbutionp P (tuple | text) of group may include:
A: determine that the probability of any theme occurs in any text；
B: determine that any theme uses the probability of any effective word；
C: determine that any theme generates the probability of any Sentiment orientation；And
D: determine that any theme results from the probability of any time.
It, can the differentiation of emotion at any time to user using the method for establishing emotion model described in the embodiment of the present application
It is predicted, compensates for the defect of the prior art.
Further, the emotion model set up according to the embodiment of the present application can also have more practical applications.
Scene one
In this scene, there is one or more text to be assessed, if needing to predict that the text to be assessed is issued,
Which type of Sentiment orientation military order reader generates.
As shown in figure 4, one embodiment of the application provides a kind of method of prediction user feeling tendency, referring to fig. 4, institute
The method of stating includes:
S400: at least one text to be assessed is determined；
S420: the model developed at any time according to a user feeling determines any at least one text to be assessed
Text enables the probability of user's generation at least Sentiment orientation.
It, can be to the user that does not deliver text and may cause using the method for prediction user feeling tendency described herein
Feelings tendency is predicted, the defect of the prior art is compensated for.
Optionally, the model that the user feeling in S420 develops at any time, can be through the application aforementioned process
(1) model that the user feeling that the method described is set up develops at any time.Training text for model foundation can be
Collection obtains in advance.
After determining one or more texts to be assessed, so that it may be located in advance to one or more of texts to be assessed
One or more of texts to be assessed, are resolved into the form of tuple above-mentioned, input the user feeling and drill at any time by reason
The model of change, and then enable user generate certain feelings as a result, obtaining any one text d ' in text to be assessed according to model output
The probability of sense tendency e, it may be assumed that P (e | d ').
Illustratively,
--- formula 1
Wherein, NtopicIndicate total quantity of theme, P (Zk| d ') indicate that theme Z occurs in text d ' to be assessedkProbability,
P(e|Zk) indicate theme ZkIt enables user produce a feeling and is inclined to the probability of e.
User feeling described in the method for emotion model is established according to the application previous embodiment at any time
The model of differentiation, it is above-mentionedIt can be provided by the output result of model.Therefore, the known variables in formula 1
For P (Zk|d′)。
According to Bayesian formula:
--- formula 2
Wherein, symbol ∝ indicates " being proportional to ", P (Zk) be the theme ZkPrior probability, which can be forefathers'
Empirical value is also possible to be obtained according to the training text during model foundation:Its
In, NdIndicate the quantity of training text.Therefore, the known variables in formula 2 be P (d ' | Zk)。
Assuming that being the model established according to the above process (1), then since assessment text d ' can be converted at least unitary
GroupForm, wherein w ' ∈ d ', therefore:
--- formula 3
Wherein, symbol ∈ indicates " belonging to ",Indicate the Sentiment orientation group that text d ' to be assessed may enable user generate
At emotion vector, t ' indicates that the release time being anticipated of text to be assessed, w ' indicate effective word of text to be assessed.
So far, all variables in formula 3 can be provided by the output result of model.The output of binding model as a result,
And above-mentioned formula 1,2,3, so that it may obtain P (e | d '), also just obtain text d ' to be assessed and user is enabled to generate certain emotion
It is inclined to the probability of e.
Scene two
In this scene, the text issued in the past period can be collected, therefrom analyzes user feeling tendency
Anomalies, and track the time that abnormal emotion tendency generates.
As shown in figure 5, a kind of method for monitoring public sentiment is provided, referring to Fig. 5, institute in another embodiment of the application
The method of stating may include:
S500: text to be assessed is determined, wherein the text to be assessed includes at least the issued in first time period
One text set and the second text set issued in second time period；
S520: the model developed at any time according to a user feeling at least determines first text set described first
Enable that user generates the probability distribution of each Sentiment orientation and second text set is enabled in the second time period period
User generates the probability distribution of each Sentiment orientation；
S540: user is enabled to generate each Sentiment orientation in the first time period according at least to first text set
Probability distribution and second text set divide in the probability that the second time period enables user generate each Sentiment orientation
Cloth determines at least abnormal emotion tendency；
S560: determine that each abnormal emotion tendency in at least abnormal emotion tendency results from any time
Probability.
Using the method for monitoring public sentiment described herein, the excavation and analysis of depth can be carried out to magnanimity public sentiment, really
Determine the differentiation of the Sentiment orientation of user at any time, compensates for the defect of the prior art.
Optionally, the model that user feeling develops at any time in the S520 can be (2) description by the above process
The model that the user feeling that method is set up develops at any time.
In this scene, the period for dividing text to be assessed is first had to, then, according to the period, to text to be assessed
Pre-processed, by the emotion of each of text vector of each text to be assessed effectively word and text to be assessed to
Amount binding, obtains the tuple of text to be assessed, inputs in submodel corresponding with the issuing time of text to be assessed itself.
According to submodel in different time periods, so that it may determine the text to be assessed of different time sections publication in the time
User is enabled to generate the probability of certain Sentiment orientation e in section:
……
Then, so that it may according to P (e | Dt1)、P(e|Dt2)、P(e|Dt3) ... determine the fluctuation situation of Sentiment orientation e.
For example, the probability distribution that the text issued in the t1 period enables user generate each Sentiment orientation can be calculated: P (e1|
Dt1), P (e2|Dt1), P (e3|Dt1), P (e4|Dt1), P (e5|Dt1) ... ... and the text issued in the t2 period enables user produce
The probability distribution of each raw Sentiment orientation: P (e1|Dt2), P (e2|Dt2), P (e3|Dt2), P (e4|Dt2), P (e5|Dt2) ... ...,
Then the distance of the two probability distribution is calculated, for example, COS distance or non symmetrical distance, thus according to above-mentioned probability distribution
Distance determine time period t 1 and time period t 2 this even in period, the fluctuation situation of each Sentiment orientation.
If the fluctuation situation of certain Sentiment orientation has been more than threshold value, so that it may think that the Sentiment orientation inclines for abnormal emotion
To.
Determine that certain abnormal emotion inclines eiTo the probability for resulting from some time, that is, determining P (t | ei):
Wherein, NtopicIndicate total quantity of theme.
Scene three
In this scene, the text that user read can be collected, the text that user read therefrom is analyzed, in emotion
Classificatory probability distribution, and service is provided based on this for user.
As shown in fig. 6, in another embodiment of the application, provides and a kind of the service of offer is inclined to based on user feeling
Method, referring to Fig. 6, the method may include:
S600: text to be assessed is determined, wherein the text to be assessed is at least text that a user read；
S620: the model developed at any time according to a user feeling determines that the text to be assessed enables the user generate
The probability distribution of each Sentiment orientation；
S640: it enables the user generate the probability distribution of each Sentiment orientation according to the text to be assessed, is described
User provides service.
Using the method described herein for being inclined to the service of offer based on user feeling, it is deep progress can be inclined to user's feelings
Degree analysis, and targetedly service is provided accordingly, compensate for the defect of the prior art.
Optionally, the model that user feeling develops at any time in the S620 can be (1) description by the above process
The model that the user feeling that method is set up develops at any time, or it is also possible to the method for (2) description by the above process
The model that the user feeling set up develops at any time.
In this scene, each text to be assessed is handled first, obtains the tuple of each text to be assessed,
The model is inputted, P (e | d ') is finally obtained.The process of P (e | d ') is obtained, has in above-mentioned scene one or scene two and retouches in detail
It states, details are not described herein again.
User is enabled to generate the probability of each Sentiment orientation according to each text to be assessed, so that it may it is all to obtain user
After the probability distribution that text to be assessed enables user generate each Sentiment orientation, optionally, it is based on the probability distribution, Ke Yiwei
The service that the user provides may include at least one of following service:
A. an at least content is pushed for the user, or request network side is that the user pushes an at least content, wherein
Obtained in probability distribution that an at least content enables the user generate each Sentiment orientation and above-mentioned S620 it is described to
The probability distribution that assessment text enables the user generate each Sentiment orientation matches.Illustratively, above-mentioned to be pushed to user
An at least content may include: an at least news；Or at least a network forum is posted；Or at least social networks hair
Speech.It is, of course, also possible to be the content of other forms, the embodiment of the present invention is not limited this.
It B. is user's commending friends, or request network side is user's commending friends, wherein the good friend reads
The probability distribution and the text to be assessed obtained in above-mentioned S620 that the text crossed enables the good friend generate each Sentiment orientation
This probability distribution for enabling the user generate each Sentiment orientation matches.
Present invention also provides a kind of devices for establishing emotion model, and referring to Fig. 7, the device for establishing emotion model can
To include:
Determining module 700, for determining an at least training text；
Preprocessing module 720 determines each training text for pre-processing to an at least training text
The emotion vector of each training text of text vector sum；
Modeling module 740, for according to the text vector of each training text, each training text
The issuing time of emotion vector and each training text determines the model that user feeling develops at any time.
Optionally, as shown in figure 8, the preprocessing module 720 may include:
Text vector pre-processes submodule 721, is filtered for the meaningless word to each training text, determines
Effective word of each training text；
Emotion vector pre-processes submodule 722, reads generate after each training text at least one for obtaining user
Sentiment orientation, and count the quantity that every kind of Sentiment orientation generates in an at least Sentiment orientation.
Optionally, as shown in figure 9, the modeling module 740 may include:
First tuple determines submodule 741, for by the effective word of each of the text vector of i-th of training text
With the issuing time binding of the emotion vector and i-th of training text of i-th of training text, i-th of training text is obtained
Tuple；
First statistic submodule 742 determines in i-th of training text for counting to i-th of training text
Occur tuple r probability P (tuple r | training text i)；
First training submodule 743, for according to the P (tuple r | training text i) is determined i-th by an algorithm
Training text select theme k probability P (theme k | training text i) and the theme k generate tuple r probability P (tuple r |
Theme k)；
First output sub-module 744, for according to the P (theme k | training text i) and the P (tuple r | theme k),
Determine that any text generates the approximate Posterior distrbutionp P (tuple | text) of any tuple.
Optionally, as shown in Figure 10, the modeling module 740 may include:
Second tuple determines submodule 745, for each of text vector of each training text is effectively single
Word and the emotion vector binding for instructing each white silk text, obtain the tuple of each training text；
Second statistic submodule 746, the training text for issuing to an at least period count, and determination is described extremely
Occur in i-th of the training text issued in a few period tuple r probability P (tuple r | training text i)；
Second training submodule 747, for according to the P (tuple r | training text i), by an algorithm, determine described in
I-th training text selection theme k probability P (theme k | training text i) and the theme k generates the probability P of tuple r
(tuple r | theme k)；
Second output sub-module 748, for according to the P (theme k | training text i) and the P (tuple r | theme k),
Determine that any text in an at least period generates the approximate Posterior distrbutionp P (tuple | text) of any tuple.
Optionally, as shown in figure 11, first output sub-module 744 or second output sub-module 748, can be with
Include:
First output unit 749-1, for determining that the probability of any theme occurs in any text；
Second output unit 749-2, for determining that any theme uses the probability of any effective word；
Third output unit 749-3, for determining that any theme generates the probability of any Sentiment orientation；And
4th output unit 749-4, for determining that any theme results from the probability of any time.
It, can the differentiation of emotion at any time to user using the device for establishing emotion model described in the embodiment of the present application
It is predicted, compensates for the defect of the prior art.
It is apparent to those skilled in the art that for convenience and simplicity of description, the foundation of foregoing description
The specific work process of the device of emotion model can refer to the correspondence of the aforementioned embodiment of the method for establishing emotion model of the application
Process description, details are not described herein.
Present invention also provides a kind of equipment of prediction user feeling tendency, as shown in figure 12, the prediction user feeling
The equipment of tendency may include:
One assessment text determining device 1200, for determining at least one text to be assessed；
One prediction meanss 1220, the model for being developed at any time according to a user feeling determine that described at least one is to be evaluated
Estimate the probability that any text in text enables user generate an at least Sentiment orientation.
Optionally, as shown in figure 13, the prediction meanss 1220 may include:
Processing module 1221 determines each text to be assessed for pre-processing at least one text to be assessed
The emotion vector of each text to be assessed of this text vector sum；
Tuple determining module 1222, for by each of text vector of each text to be assessed effectively word and
The binding of the issuing time of the emotion vector of each text to be assessed and each text to be assessed, it is to be assessed to obtain each
The tuple of text；
Input module 1223 is drilled at any time for the tuple of each text to be assessed to be inputted the user feeling
The model of change, the output of the model developed at any time according to the user feeling is as a result, determine at least one text to be assessed
In any text enable user generate an at least Sentiment orientation probability.
Optionally, as shown in figure 14, the equipment of the prediction user feeling tendency can also include:
One establishes the device 1240 of emotion model, the model developed at any time for establishing the user feeling.
Optionally, as shown in figure 15, the device 1240 for establishing emotion model may include:
Determining module 1241, for determining an at least training text；
Preprocessing module 1242 determines each training text for pre-processing to an at least training text
Text vector sum each training text emotion vector；
Modeling module 1243, for the institute according to the text vector of each training text, each training text
Emotion vector and the issuing time of each training text are stated, determines the model that user feeling develops at any time.
Optionally, the preprocessing module 1242 can be the preprocessing module 720 in embodiment as shown in Figure 8, herein
It repeats no more.
Optionally, the modeling module 1243 can be the modeling module 740 in embodiment as shown in Figure 9, herein no longer
It repeats.
It, can be to the user that does not deliver text and may cause using the equipment of prediction user feeling tendency described herein
Sentiment orientation is predicted, the defect of the prior art is compensated for.
Present invention also provides a kind of equipment for monitoring public sentiment, and as shown in figure 16, the equipment of the monitoring public sentiment can wrap
It includes:
One assessment text determining device 1600, for determining text to be assessed, wherein the text to be assessed includes at least
The first text set issued in first time period and the second text set issued in second time period；
One emotion probability distribution determining device 1620, the model for being developed at any time according to a user feeling, at least really
Fixed first text set enables user generate the probability distribution and described the of each Sentiment orientation in the first time period
Two text sets enable user generate the probability distribution of each Sentiment orientation in the second time period；
One abnormal emotion determining device 1640 is used for enabling according at least to first text set in the first time period
Family generate the probability distribution of each Sentiment orientation and second text set to enable user generate in the second time period every
A kind of probability distribution of Sentiment orientation determines at least abnormal emotion tendency；
One time probability distribution determining device 1660, it is different for each in the determining at least abnormal emotion tendency
Normal Sentiment orientation results from the probability of any time.
Optionally, as shown in figure 17, the emotion probability distribution determining device 1620 may include:
Processing module 1621 determines the text of each text to be assessed for pre-processing to the text to be assessed
The emotion vector of each text to be assessed of word vector sum；
Tuple determining module 1622, for by each of text vector of each text to be assessed effectively word and
The emotion vector of each text to be assessed is bound, and the tuple of each text to be assessed is obtained；
Input module 1623, for the tuple of each text to be assessed to be inputted what the user feeling developed at any time
Submodel corresponding with the issuing time of each text to be assessed, develops at any time according to the user feeling in model
Model output as a result, at least determining that first text concentrates on the first time period and user is enabled to generate each emotion
The probability distribution of tendency and second text set enable user generate the general of each Sentiment orientation in the second time period
Rate distribution.
Optionally, as shown in figure 18, the abnormal emotion determining device 1640 may include:
Sentiment orientation fluctuates determining module 1641, at least determining that the text issued in first time period enables user generate
The probability distribution of each Sentiment orientation enables user generate the general of each Sentiment orientation with the text issued in second time period
The distance of rate distribution；
Abnormal emotion determining module 1642, it is each for enabling user generate according to the text issued in the first time period
The probability that the text issued in the probability distribution and second time period of kind Sentiment orientation enables user generate each Sentiment orientation divides
The distance of cloth determines at least abnormal emotion tendency.
Optionally, as shown in figure 19, the equipment of the monitoring public sentiment can also include:
One establishes the device 1680 of emotion model, the model developed at any time for establishing the user feeling.
Optionally, as shown in figure 20, the device 1680 for establishing emotion model may include:
Determining module 1681, for determining an at least training text；
Preprocessing module 1682 determines each training text for pre-processing to an at least training text
Text vector sum each training text emotion vector；
Modeling module 1683, for the institute according to the text vector of each training text, each training text
Emotion vector and the issuing time of each training text are stated, determines the model that user feeling develops at any time.
Optionally, the preprocessing module 1682 can be the preprocessing module 720 in embodiment as shown in Figure 8, herein
It repeats no more.
Optionally, the modeling module 1683 can be the modeling module 740 in embodiment as shown in Figure 10, herein no longer
It repeats.
Using the equipment of monitoring public sentiment described herein, the excavation and analysis of depth can be carried out to magnanimity public sentiment, really
Determine the differentiation of the Sentiment orientation of user at any time, compensates for the defect of the prior art.
One embodiment of the application additionally provides a kind of equipment for being inclined to the service of offer based on user feeling, referring to figure
21, it is described based on user feeling be inclined to provide service equipment may include:
One assessment text determining device 2100, for determining text to be assessed, wherein the text to be assessed is a user
At least text read；
One emotion probability distribution determining device 2120, the model for being developed at any time according to a user feeling determine institute
State the probability distribution that text to be assessed enables the user generate each Sentiment orientation；
One service providing apparatus 2140, for enabling the user generate each Sentiment orientation according to the text to be assessed
Probability distribution, provide service for the user.
Optionally, as shown in figure 22, the emotion probability distribution determining device 2120 may include:
First processing module 2121 determines each text to be assessed for pre-processing to the text to be assessed
Text vector sum each text to be assessed emotion vector；
First tuple determining module 2122, for each of text vector of each text to be assessed is effectively single
The issuing time binding of the emotion vector and each text to be assessed of word and each text to be assessed, obtains each and waits for
Assess the tuple of text；
First input module 2123 is drilled at any time for the tuple of each text to be assessed to be inputted the user feeling
The model of change, the output of the model developed at any time according to the user feeling is as a result, determine that the text to be assessed enables user
Generate the probability distribution of each Sentiment orientation.
Optionally, as shown in figure 23, the emotion probability distribution determining device 2120 may include:
Second processing module 2124 determines each text to be assessed for pre-processing to the text to be assessed
Text vector sum each text to be assessed emotion vector；
Second tuple determining module 2125, for each of text vector of each text to be assessed is effectively single
The emotion vector of word and each text to be assessed is bound, and the tuple of each text to be assessed is obtained；
Second input module 2126 is drilled at any time for the tuple of each text to be assessed to be inputted the user feeling
Submodel corresponding with the issuing time of each text to be assessed, develops at any time according to the user feeling in the model of change
Model output as a result, determining the probability distribution that the text to be assessed enables user generate each Sentiment orientation.
Optionally, as shown in figure 24, the service providing apparatus 2140 may include:
First service provides module 2141, for being that the user pushes an at least content, or request network side is described
User pushes an at least content, wherein the probability distribution that an at least content enables the user generate each Sentiment orientation
The probability distribution for enabling the user generate each Sentiment orientation with the text to be assessed matches；And/or
Second service provides module 2142, for pushing away for user's commending friends, or request network side for the user
Recommend good friend, wherein the text that the good friend read enable the good friend generate the probability distribution of each Sentiment orientation with it is described
The probability distribution that text to be assessed enables the user generate each Sentiment orientation matches.
Optionally, as shown in figure 25, the equipment for being inclined to the service of offer based on user feeling can also include:
One establishes the device 2160 of emotion model, the model developed at any time for establishing the user feeling.
Optionally, the device 2160 for establishing emotion model can be the implementation as shown in Fig. 7 either figure into Figure 11
The device of emotion model is established described in example, details are not described herein again.
Using the equipment described herein for being inclined to the service of offer based on user feeling, it is deep progress can be inclined to user's feelings
Degree analysis, and targetedly service is provided accordingly, compensate for the defect of the prior art.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and method and step can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
Scope of the present application.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, controller or network equipment etc.) execute each embodiment the method for the application all or part of the steps.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
Embodiment of above is merely to illustrate the application, and is not the limitation to the application, in relation to the common of technical field
Technical staff can also make a variety of changes and modification in the case where not departing from spirit and scope, therefore all
Equivalent technical solution also belongs to the scope of the application, and the scope of patent protection of the application should be defined by the claims.
Claims (4)
1. a kind of method for being inclined to the service of offer based on user feeling characterized by comprising
Determine text to be assessed, wherein the text to be assessed is at least text that a user read；
According to the model that a user feeling develops at any time, determine that the text to be assessed enables the user generate each emotion
The probability distribution of tendency, wherein the model that the user feeling develops at any time pre-establishes；
It enables the user generate the probability distribution of each Sentiment orientation according to the text to be assessed, provides clothes for the user
Business；
Wherein, the model that the user feeling develops at any time is established, comprising:
Determine an at least training text；
An at least training text is pre-processed, determines each training text of the text vector sum of each training text
This emotion vector；
According to the text vector of each training text, the emotion vector of each training text and each instruction
The issuing time for practicing text, determines the model that user feeling develops at any time；
The probability distribution for enabling the user generate each Sentiment orientation according to the text to be assessed mentions for the user
For service, comprising:
For the user push an at least content, or request network side be the user push an at least content, wherein it is described extremely
A few content enables that the user generates the probability distribution of each Sentiment orientation and the text to be assessed enables user's generation
The probability distribution of each Sentiment orientation matches；And/or
For user's commending friends, or request network side is user's commending friends, wherein the text that the good friend read
This enables that the good friend generates the probability distribution of each Sentiment orientation and the text to be assessed enables the user generate each
The probability distribution of Sentiment orientation matches.
2. the method as described in claim 1, which is characterized in that according to the model that a user feeling develops at any time, determine institute
State the probability distribution that text to be assessed enables the user generate each Sentiment orientation, comprising:
The text to be assessed is pre-processed, determines each text to be assessed of the text vector sum of each text to be assessed
This emotion vector；
The emotion of each of text vector by each text to be assessed effectively word and each text to be assessed to
Amount and the binding of the issuing time of each text to be assessed, obtain the tuple of each text to be assessed；
The tuple of each text to be assessed is inputted into the model that the user feeling develops at any time, according to the user feeling
The output of the model developed at any time is as a result, determine the probability point that the text to be assessed enables user generate each Sentiment orientation
Cloth；
Or
The text to be assessed is pre-processed, determines each text to be assessed of the text vector sum of each text to be assessed
This emotion vector；
The emotion of each of text vector by each text to be assessed effectively word and each text to be assessed to
Amount binding, obtains the tuple of each text to be assessed；
The tuple of each text to be assessed is inputted to be assessed with each in the model that the user feeling develops at any time
The corresponding submodel of the issuing time of text, the output of the model developed at any time according to the user feeling is as a result, determine institute
State the probability distribution that text to be assessed enables user generate each Sentiment orientation.
3. a kind of equipment for being inclined to the service of offer based on user feeling, which is characterized in that described to be provided based on user feeling tendency
The equipment of service includes:
One assessment text determining device, for determining text to be assessed, wherein the text to be assessed is that a user read
An at least text；
One emotion probability distribution determining device, the model for being developed at any time according to a user feeling determine described to be assessed
Text enables the user generate the probability distribution of each Sentiment orientation；
One service providing apparatus, the probability for enabling the user generate each Sentiment orientation according to the text to be assessed divide
Cloth provides service for the user；
One establishes the device of emotion model, the model developed at any time for establishing the user feeling；
The device for establishing emotion model includes:
Determining module, for determining an at least training text；
Preprocessing module, for being pre-processed to an at least training text, determine the text of each training text to
The emotion vector of amount and each training text；
Modeling module, for according to the emotion of the text vector of each training text, each training text to
The issuing time of amount and each training text, determines the model that user feeling develops at any time；
The service providing apparatus includes:
First service provides module, for pushing an at least content, or request network side for the user as user push
An at least content, wherein probability distribution that an at least content enables the user generate each Sentiment orientation and it is described to
The probability distribution that assessment text enables the user generate each Sentiment orientation matches；And/or
Second service provides module, is used to be user's commending friends, or request network side is user's commending friends,
In, the text that the good friend read enable the good friend generate each Sentiment orientation probability distribution and the text to be assessed
The probability distribution for enabling the user generate each Sentiment orientation matches.
4. equipment as claimed in claim 3, which is characterized in that the emotion probability distribution determining device includes:
First processing module, for being pre-processed to the text to be assessed, determine the text of each text to be assessed to
The emotion vector of amount and each text to be assessed；
First tuple determining module, for by each of text vector of each text to be assessed effectively word and each
The issuing time binding of the emotion vector and each text to be assessed of a text to be assessed, obtains each text to be assessed
Tuple；
First input module, for the tuple of each text to be assessed to be inputted the mould that the user feeling develops at any time
Type, the output of the model developed at any time according to the user feeling is as a result, determine that the text to be assessed enables user generate often
A kind of probability distribution of Sentiment orientation；
Or
Second processing module, for being pre-processed to the text to be assessed, determine the text of each text to be assessed to
The emotion vector of amount and each text to be assessed；
Second tuple determining module, for by each of text vector of each text to be assessed effectively word and each
The emotion vector of a text to be assessed is bound, and the tuple of each text to be assessed is obtained；
Second input module, for the tuple of each text to be assessed to be inputted the model that the user feeling develops at any time
In submodel corresponding with the issuing time of each text to be assessed, the model developed at any time according to the user feeling
Output is as a result, determine the probability distribution that the text to be assessed enables user generate each Sentiment orientation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410773678.3A CN104504032B (en) | 2014-12-12 | 2014-12-12 | The method and apparatus for being inclined to the service of offer based on user feeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410773678.3A CN104504032B (en) | 2014-12-12 | 2014-12-12 | The method and apparatus for being inclined to the service of offer based on user feeling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104504032A CN104504032A (en) | 2015-04-08 |
| CN104504032B true CN104504032B (en) | 2019-03-01 |
Family
ID=52945430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410773678.3A Active CN104504032B (en) | 2014-12-12 | 2014-12-12 | The method and apparatus for being inclined to the service of offer based on user feeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104504032B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460111B (en) * | 2018-02-08 | 2020-10-16 | 合肥工业大学 | Personal character modeling and generating method and device based on conversation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103365867A (en) * | 2012-03-29 | 2013-10-23 | 腾讯科技（深圳）有限公司 | Method and device for emotion analysis of user evaluation |
| CN103559233A (en) * | 2012-10-29 | 2014-02-05 | 中国人民解放军国防科学技术大学 | Extraction method for network new words in microblogs and microblog emotion analysis method and system |
| TW201416887A (en) * | 2012-10-30 | 2014-05-01 | Univ Ming Chuan | Methods for sentimental analysis of news text |
| CN104063399A (en) * | 2013-03-22 | 2014-09-24 | 杭州金弩信息技术有限公司 | Method and system for automatically identifying emotional probability borne by texts |
-
2014
- 2014-12-12 CN CN201410773678.3A patent/CN104504032B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103365867A (en) * | 2012-03-29 | 2013-10-23 | 腾讯科技（深圳）有限公司 | Method and device for emotion analysis of user evaluation |
| CN103559233A (en) * | 2012-10-29 | 2014-02-05 | 中国人民解放军国防科学技术大学 | Extraction method for network new words in microblogs and microblog emotion analysis method and system |
| TW201416887A (en) * | 2012-10-30 | 2014-05-01 | Univ Ming Chuan | Methods for sentimental analysis of news text |
| CN104063399A (en) * | 2013-03-22 | 2014-09-24 | 杭州金弩信息技术有限公司 | Method and system for automatically identifying emotional probability borne by texts |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104504032A (en) | 2015-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Medvet et al. | Publication venue recommendation based on paper abstract | |
| Nigam et al. | Towards a robust metric of opinion | |
| CN108108821A (en) | Model training method and device | |
| Meraz et al. | Networked framing and gatekeeping | |
| JP5945062B2 (en) | Data processing apparatus and story model construction method | |
| CN106780073B (en) | Social network influence maximization initial node selection method considering user behaviors and emotions | |
| CN113748439B (en) | Prediction of successful quotient of movies | |
| Yang et al. | User interest and social influence based emotion prediction for individuals | |
| US10496751B2 (en) | Avoiding sentiment model overfitting in a machine language model | |
| CN103294664A (en) | Method and system for discovering new words in open fields | |
| CN108845986A (en) | A kind of sentiment analysis method, equipment and system, computer readable storage medium | |
| CN105069036A (en) | Information recommendation method and apparatus | |
| Dong et al. | Further experiments in opinionated product recommendation | |
| CN106407381A (en) | Method and device for pushing information based on artificial intelligence | |
| CN112949973A (en) | AI-combined robot process automation RPA process generation method | |
| Ogudo et al. | Sentiment analysis application and natural language processing for mobile network operators’ support on social media | |
| KR101931624B1 (en) | Trend Analyzing Method for Fassion Field and Storage Medium Having the Same | |
| EP2613275B1 (en) | Search device, search method, search program, and computer-readable memory medium for recording search program | |
| El Morr et al. | Virtual community building and the information society: Current and future directions | |
| KR101575779B1 (en) | Program rating prediction method and apparatus, and system based on sentiment analysis of viewers comments | |
| JP5973935B2 (en) | Browsing behavior prediction device, browsing behavior prediction method, and program | |
| Nirmala et al. | Twitter data analysis for unemployment crisis | |
| Susilawati | Public services satisfaction based on sentiment analysis: Case study: Electrical services in Indonesia | |
| CN104484437A (en) | Network brief comment sentiment mining method | |
| CN104516873A (en) | Method and device for building emotion model |
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 |