CN110134794A - A kind of construction method and device of entity portrait - Google Patents
A kind of construction method and device of entity portrait Download PDFInfo
- Publication number
- CN110134794A CN110134794A CN201910308951.8A CN201910308951A CN110134794A CN 110134794 A CN110134794 A CN 110134794A CN 201910308951 A CN201910308951 A CN 201910308951A CN 110134794 A CN110134794 A CN 110134794A
- Authority
- CN
- China
- Prior art keywords
- entity
- label
- keyword
- iteration
- weight
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Animal Behavior & Ethology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the present application discloses the construction method and device of a kind of entity portrait, and method includes: to determine the corresponding keyword of the entity according to the text description information of the entity；According to the feature vector of entity described in browsing behavior information acquisition of the user to the entity；Label extraction is carried out based on the keyword and described eigenvector, obtains label；The corresponding label portrait of the entity is constructed according to the label of extraction.The embodiment of the present application is suitable for the entity portrait building of short text and the sparse scene of corpus, meets application demand, and be based on label depicting entity, facilitates observation industry overall picture and industry details.
Description
Technical field
The invention relates to computer internet technical fields, and in particular to a kind of construction method of entity portrait and
Device.
Background technique
Currently, entity portrait building is based primarily upon Text Feature Extraction algorithm, i.e., closed by the text description information of entity
Keyword, and then entity portrait is constructed, but the program is not suitable for short text or the sparse situation of corpus, because of short text or language
Expect it is sparse lack enough description informations, be difficult to realize only according to limited text information to the building of the portrait of entity, and more
Entity all has apparent long tail effect in number situation, there are corpus Sparse Problems, the entity constructed hereby based on Text Feature Extraction
Portrait is not comprehensive, accuracy is lower.
Summary of the invention
In view of this, the embodiment of the present application provides the construction method and device of a kind of entity portrait, described based on text
Information and user behavior information extraction label solve text extraction and portray entity caused by entity according to label depicting entity
The technical problem that portrait is not comprehensive, accuracy is not high.
According to the first aspect of the application, a kind of construction method of entity portrait is provided, comprising:
The corresponding keyword of the entity is determined according to the text description information of the entity；
According to the feature vector of entity described in browsing behavior information acquisition of the user to the entity；
Label extraction is carried out based on the keyword and described eigenvector, obtains label；
The corresponding label portrait of the entity is constructed according to the label of extraction.
According to the second aspect of the application, a kind of construction device of entity portrait is provided, comprising:
Label abstraction module determines the corresponding keyword of the entity for the text description information according to the entity；
According to the feature vector of entity described in browsing behavior information acquisition of the user to the entity；Based on the keyword and described
Feature vector carries out label extraction, obtains label；
Portrait building module is drawn a portrait for constructing the corresponding label of the entity according to the label of extraction.
In terms of according to the third of the application, a kind of electronic equipment is provided, electronic equipment includes: processor, Yi Jicun
Contain the memory for the computer program that can be run on a processor；Wherein, the processor, for executing the memory
In computer program when execute method described in the application first aspect.
According to the 4th of the application the aspect, a kind of computer readable storage medium is provided, computer journey is stored thereon with
Sequence, the computer program realize method described in the application first aspect when being executed by processor.
Using construction method and device that the entity of the embodiment of the present application is drawn a portrait, determined according to the text description information of entity
The corresponding keyword of entity, according to user to the feature vector of the browsing behavior information acquisition entity of entity；Based on keyword with
And feature vector carries out label and extracts to obtain label；The corresponding label portrait of entity is constructed according to the label of extraction.This Shen as a result,
Please not only consider the text description information of entity but also the browsing behavior information extraction label based on user to entity, solves portion
The text description information missing of point entity be difficult to construct entity user's portrait of portrait or building not comprehensively, accuracy it is not high
Technical problem.
Detailed description of the invention
Fig. 1 is the flow chart of the construction method of the entity portrait of the application one embodiment；
Fig. 2 is the schematic diagram of the application one embodiment tag extraction；
Fig. 3 is the schematic diagram that the application one embodiment is iterated label；
Fig. 4 is the block diagram of the construction device of the application one embodiment entity portrait；
Fig. 5 is the structural schematic diagram of the electronic equipment of the application one embodiment；
Fig. 6 is the structural schematic diagram of the computer readable storage medium of the application one embodiment.
Specific embodiment
To keep the above objects, features, and advantages of the embodiment of the present application more obvious and easy to understand, with reference to the accompanying drawing and
Specific embodiment is described in further detail the embodiment of the present application.Obviously, described embodiment is that the application is implemented
Example a part of the embodiment, instead of all the embodiments.Based on the embodiment in the embodiment of the present application, ordinary skill people
Member's every other embodiment obtained without making creative work, belongs to the model of the embodiment of the present application protection
It encloses.
Fig. 1 is the flow chart of the construction method of the entity portrait of the application one embodiment, referring to Fig. 1, the present embodiment
The construction method of entity portrait includes the following steps:
Step S101 determines the corresponding keyword of the entity according to the text description information of the entity；
Step S102, according to the feature vector of entity described in browsing behavior information acquisition of the user to the entity；
Step S103 carries out label extraction based on the keyword and described eigenvector, obtains label；
Step S104 constructs the corresponding label of the entity according to the label of extraction and draws a portrait.
As shown in Figure 1 it is found that the construction method of entity portrait through this embodiment, realizes and portray entity automatically.And
And it is real according to label depicting in turn to the browsing behavior information excavating and extraction label of entity based on text description information and user
Body solves the technical problem that the portrait of user caused by short text in the prior art and Sparse is not comprehensive, accuracy is lower,
Improve the accuracy of the entity portrait of building.
It should be noted that foregoing schemes can be in the calculating of such as a group of computer-executable instructions the step of illustrating
It is executed in machine system, although also, logical order is shown in flow charts, and it in some cases, can be to be different from
Sequence herein executes shown or described step.For example, the sequence of abovementioned steps S101 and step S102 can be interchanged, i.e.,
Step S102 is first carried out, step S101 is then executed.The entity of the present embodiment includes trade company, and trade company here can be occupancy object
The entity trade company for managing space can also be without limitation with Xian Shang trade company.
When specific implementation, the construction method of the entity portrait of the present embodiment obtains information first from data source, acquisition
Information includes two kinds, and one is users for the browsing behavior information of entity, another is text information, such as the text of entity
This description information.After getting information, label extraction is carried out.Which keyword that is mainly used for judging that label extracts can be with
As label.Label portrait can be obtained to entity progress label depicting later by being drawn into label.
It is that entity is drawn a portrait that so-called label portrait, which is based on label, that is, determine multiple labels that each entity includes,
Label is the signature identification of entity, for example the label of entity KFC can be the expression such as chain fast food, fried chicken, hamburger entity
The keyword of main business.
Since label extraction is the basis of label portrait building, first label extraction is illustrated below.
The label portrait construction method of the present embodiment is extracted and is marked based on keyword and feature vector progress label
Label, label is carried out based on keyword and feature vector and extracts to obtain label to include: the feature vector based on entity, calculates and appoints
The similarity anticipated between two entities；Similarity network is constructed according to similarity；It is right to obtain first instance institute in similarity network
Second keyword corresponding to each neighboring entities of the first keyword and first instance answered, by the first keyword and second
Keyword obtains keyword set；To in keyword set keyword carry out label extraction, using the keyword extracted as
Label.
It has been observed that label is to carry out label based on keyword and feature vector to extract, i.e. label extraction includes
Keyword is determining and feature vector determines two sub-steps, and feature vector here is believed according to browsing behavior of the user to entity
What breath obtained, and it is specifically included according to browsing behavior information acquisition feature vector of the user to entity: according to browsing behavior information,
Target group's index TGI value between any two entity is obtained, entity is encoded, and the TGI value of entity is suitable by encoding
Sequence is arranged, and the feature vector of entity is obtained.
TGI (Target Group Index, target group's index) is for measuring a group relative to another
The tendentiousness of group, for example, TGI (entity A entity B) be equal in the user of browsing entity A also browsed user's accounting of entity B/
User's accounting of entity B has been browsed in standard group.Symbol "/" indicate divided by.
Here according to the browsing behavior information, the target group's index TGI value obtained between any two entity includes:
According to the browsing behavior information of first instance and the browsing behavior information of second instance, the user for having browsed first instance is determined
In also browsed the first number corresponding to the user of second instance, calculate the first number and browsed the number of users of first instance
Purpose ratio obtains the first accounting；The second number according to corresponding to the user for having browsed second instance in the standard group of acquisition
Mesh calculates the ratio of the second number and the number of standard group, obtains the second accounting；It is accounted for by first accounting with described second
The ratio of ratio obtains target group's index TGI value between first instance and second instance.
For example, the number of users for having browsed entity A (or first instance) is 100 (only making example), in this 100 users
In there are 10 users also to browse entity B (or second instance), then browse the user that entity B has also been browsed in the user of entity A
Accounting is 10%.The number of users of standard group is 500 (only making example), and the user that entity B has been browsed in standard group is 200,
User's accounting that entity B has then been browsed in standard group is 40%, by the use for also having browsed entity B in the user of browsing entity A
User's accounting 40% of entity B is browsed in family accounting 10% and standard group, use 10% can be obtained divided by 40%, and TGI is (real
Body A → entity B)=25%, that is, target group's index TGI value of entity A and entity B is 25%.And so on, it can calculate
The tendentiousness of any two entity in the entity got out.
In view of the difference for the user group that Xian Shang trade company and Xian Xia trade company face, for the quotient on line in the present embodiment
Family determines the number of users of standard group according to all users in scope of statistics.And for Xian Xia trade company, then according to preset distance
Interior user determines the number of users of standard group.For example, for Xian Xia trade company KFC, according to the geographical location of the user of acquisition
Data are user that all numbers of users in radius are determined as standard group by being the center of circle with 5 kilometers with entity (such as KFC)
Number.
Different from traditional with the latent structures scheme such as category, sales volume, price, according to the TGI value of entity in the present embodiment
The feature vector for constructing entity, after specifically calculating target group's index TGI value between first instance and second instance,
(coded sequence is random) is encoded to entity, the TGI value by target entity relative to other entities is carried out according to coded sequence
Arrangement forms vector, which is the feature representation namely feature vector of entity.For example, in one embodiment, entity
Quantity is that 5 (only making example) are a, according to preceding description, calculates tendentiousness i.e. target group's index TGI of any two entity
Value, can obtain the TGI value of every 1 entity and remaining 4 entity.After obtaining these TGI values, entity is encoded, such as
It is sequentially entity 1, entity 2, entity 3, entity 4 and entity 5 after coding, then the feature vector 1 of entity 1 is by [entity 1 is opposite
In the TGI value of entity 2, TGI value of the entity 1 relative to entity 3, TGI value of the entity 1 relative to entity 4, entity 1 is relative to reality
The TGI value of body 5] composition.Similarly, the feature vector 2 of entity 2 be by [TGI value of the entity 2 relative to entity 1, entity 2 relative to
The TGI value of entity 3, TGI value of the entity 2 relative to entity 4, TGI value of the entity 2 relative to entity 5] composition, and so on, it can
Entity 3, entity 4, the feature vector of entity 5 are obtained respectively.The substance feature design of the present embodiment can express a reality as a result,
Body distinguishes the information with other entities, and then the entity portrait precision constructed is higher.
Keyword in aforementioned label extraction is key determining according to the text description information of entity, corresponding with entity
Word.Determine that keyword corresponding with entity includes: the text based on tokenizer to entity according to the text description information of entity
Description information carries out text word cutting, such as " seafood, the quantity-unlimiting supply of sashimi " becomes " seafood ", " sashimi ", " no after word cutting
Limitation " " supply ".Then preliminary screening is carried out to obtained word cutting, preliminary screening here is filtered including 1) word frequency, to all
Word cutting carry out word frequency statistics, word frequency is filtered out lower than the word of given threshold.2) part of speech filters, and carries out to complementary word
Filter.3) stop words filters, as " ", " I " meaningless word is filtered.It is corresponding that each entity is obtained after preliminary treatment
Keyword.
Since the text description information that entities a large amount of in practice all have corpus Sparse Problems or entity is short text, thus
It is difficult to construct accurate, comprehensive entity portrait, in this regard, proposing using the label of neighboring entities in the present embodiment to target entity
Label is expanded.Specifically, on the basis of the substance feature vector that previous constructions go out, it is similar according to cosine in the present embodiment
The similarity that formula calculates any two entity is spent, similarity network is constructed according to similarity.
Cosine similarity is the similarity that them are assessed by calculating the included angle cosine value of two vectors, specific to this Shen
It please be feature vector (feature vector 1, the feature for calculating every two entity (such as aforementioned entities 1 and aforementioned entities 2) in embodiment
Vector 2) between the cosine value of angle determine the similarities of two entities.Any two entity is calculated based on cosine similarity formula
Similarity be the prior art, may refer to record in the prior art, do not do excessive explanation here.
After the similarity (i.e. cosine value) of any two entity is calculated, similarity and preset threshold are compared
Compared with if similarity is greater than the preset threshold, then it is assumed that have side connection between two entities, otherwise it is assumed that connecting without side, such as
There is side connection between two entities of fruit, side right weight is the similarity of two entities.
As shown in Fig. 2, including multiple entities in the similarity network of the present embodiment, wherein each App represents a reality
Body, there is side connection between part entity, such as number 0.8 representative edge weight namely entity of the App2 and App4 there are side, on side
The similarity of App2 and App4.
In the present embodiment, for any entity, by the keyword of entity itself in the similarity network of aforementioned building (such as
1 keyword " lodging "), and keyword corresponding with the entity that target entity has side to connect (such as 2 keyword " wine
Shop ", " people place ") composition keyword set (keyword set include 3 keywords, respectively be stay ", " hotel ", " people place ").
That is, being the keyword of multiple entity by the keyword expansion of single entities, the description information for thus solving part entity is short essay
This or corpus is sparse is difficult to extract crucial word problem.
Referring to fig. 2, the right half part of Fig. 2 illustrates a keyword set, includes four passes in the keyword set
Keyword is Word1, Word2, Word3 and Word4 respectively.In this four keywords, Word2 and Word3 are highlighted
(font size is bigger than the font size of other 2 keywords) is because the two keywords are core word, i.e. TF-IDF is worth dividing high
Word, TF-IDF are the theme score, and value is higher to show that keyword is more important, it is known that core word is that those are heavier in keyword set
The word wanted, that is, can be used as the word of label.By can be obtained core word combination all in keyword set each
The final tag set of entity.
Specifically after obtaining keyword set, the present embodiment carries out label pumping to the keyword in keyword set
It takes.Label extraction is carried out to the keyword in keyword set to specifically include: obtaining the TF- of each keyword in keyword set
The keyword abstraction that TF-IDF score is greater than default score threshold is come out and is used as label by IDF score.
TF-IDF (Term Frequency-Inverse Document Frequency) be it is a kind of for information retrieval with
The common weighting technique of data mining.TF refers to word frequency, and IDF refers to inverse text frequency, and TF can be understood as keyword set
In certain keyword weighting frequency, IDF are reverse frequency, the i.e. opposite number of the logarithm of keyword overall distribution probability.Meter
TF-IDF value is greater than the keyword of given threshold as label by the TF-IDF value for calculating each keyword.
The above are the whole process that label extracts, and the process of building label portrait is described below.
In the present embodiment, the label of entity is excavated using label propagation algorithm.Different from traditional entity tag side of portraying
Formula, the scheme of the present embodiment is in addition to it can be found that other than the feature of entity itself, additionally it is possible to spy that discovery is more extended, expanding
Sign is to be promoted to the rich of entity description.For example, some trade company goes back in addition to that can excavate " coffee shop " this foundation characteristic
It can be found that the deeper characteristic information such as " literature and art ", " nourishment for the mind ", to more precisely portray entity.
Specifically, constructing the corresponding label portrait of entity according to the label of extraction includes: to initialize in similarity network in fact
The label of body obtains the tally set of each entity, includes N number of label in tally set, N is natural number；The tally set of each entity is traversed,
Label in the tally set is iterated, the corresponding mark of the entity is constructed according to the tally set of entity each when stopping iteration
Label portrait；Label in the tally set is iterated and is specifically included: according to weight of the label on first instance, the mark
The side right weight of the weight and the first instance and the neighboring entities on neighboring entities is signed, is obtained described in epicycle iteration
New weight after tag update, and the expansion label in the tally set of the neighboring entities is added to the institute of the first instance
It states and obtains new tally set in tally set, wherein described expand in the tally set that label is the neighboring entities is different from described the
Label in the tally set of one entity；New weight after the new weight and last round of iteration of label after calculating epicycle iteration
Variable quantity, according to the comparison result of the variable quantity and default iteration threshold, it is determined whether stop iteration, when stopping iteration
Determine the tally set of each entity.
It includes: by the institute of extraction that the label of entity, which obtains the tally set of each entity, in similarity network described in aforementioned initialization
It states label to be matched with the keyword in the text description information of each entity, if matching is consistent, by matched pass
Keyword is added in sky tally set as entity from tape label, to obtain the tally set of each entity.
That is, the present embodiment first initializes similarity network when constructing label portrait.Initialization is
Entity is regard as initial labels from tape label (label obtained i.e. from text information based on text extraction algorithm).Included mark
Label are to obtain keyword after carrying out the processing such as word cutting, preliminary screening by text information to entity, by keyword and label into
Row matching, since label is that TF-IDF value is greater than the keyword of given threshold so the two can be matched directly, if closed
Keyword has matched label, for example, keyword is to stay, label is also to stay, it is determined that keyword is that one of entity is included
Label.
It is for extension tag from tape label, extension tag is that the label of other entity extends to current reality
Body, why being extended to label is because some entity text description informations lack, it is difficult to obtain valuable label, nothing
Method is precisely drawn a portrait.For each entity, sky tally set is allocated in advance, if sporocarp has from tape label, then will add from tape label
Into empty tally set, the tally set of each entity is obtained.
After the tally set for obtaining each entity, in order to which the label in the tally set to entity is extended, the present embodiment
It is middle to use label propagation algorithm, the tally set of each entity is traversed, the label in tally set is iterated, when according to stopping iteration
The tally set of each entity constructs the corresponding label portrait of the entity.
Here, the label in tally set is iterated and is specifically included: according to weight of the label on first instance, mark
The side right weight of the weight and first instance and neighboring entities on neighboring entities is signed, it is updated to obtain label in epicycle iteration
New weight, and the expansion label in the tally set of neighboring entities is added in the tally set of first instance and obtains new tally set,
The new weight of epicycle label and the variable quantity of last round of new weight are calculated after a wheel iteration, according to variable quantity and is preset repeatedly
For the comparison result of threshold value, it is determined whether stop iteration, the tally set of each entity is determined when stopping iteration.
Referring to Fig. 3, the label primary iteration i.e. process of first round iteration is illustrated.It mainly includes three steps, respectively
It is:
Weight of the step (1) according to label on first instance, weight and first instance of the label on neighboring entities
With the side right weight of neighboring entities, the updated new weight of label in epicycle iteration is obtained.
Referring to Fig. 3, Fig. 3 illustrate in a similarity network include four entities the case where, four entities are real respectively
Body A, entity B, entity C and entity D.Left side illustrates label right side signal in weight physically, Fig. 3 before iteration in Fig. 3
The result of label weight variation and tag extension after iteration.
Here first the variation of label weight is illustrated, by taking " health " label of entity A as an example.
In order to facilitate calculating, it is 1 that the present embodiment, which sets all labels in initial weight physically when initializing,.And lead to
Cross the new weight that following equation calculates label after a wheel iteration:
W (new)=(old weight of the 0.5 × label on first instance) (label is on each neighboring entities by+0.5 × sum
Old weight × side right weight).
Wherein, 0.5 is constant, and the label weight of expression current entity and neighboring entities respectively takes the meaning of half.
As shown in figure 3, the initial weight of " health " label is 1.0, that is, healthy label is in first instance (entity A)
Old weight is 1.0, since the neighboring entities of entity A are entity B and entity C, does not have " health " label on entity B and entity C,
So value of the neighbor AP P on certain label (i.e. healthy label) is 0, the side right weight of entity A and neighboring entities B are 0.5, entity A
Be 0.9 with the side right weight of neighboring entities C, according to above-mentioned formula calculate new weight W (new)=(0.5 × 1.0) of " health " label+
0.5 × sum (0 × 0.5+0 × 0.9)=0.5.
According to above-mentioned identical step, remaining 2 for can calculating entity A newly weigh from tape label " checking card " and " net is red "
It is again respectively 0.75 and 0.5.
It can similarly obtain, the new weight from tape label " high face value ", " dining room ", " checking card " and " appointment " of entity B is respectively:
0.5,0.5,0.75,0.5.The new weight from tape label " dining room " and " no peppery not joyous " of entity D is respectively: 0.5 and 0.5.
For convenient data processing, label weight is mapped in 0~1 range so that handling more just in the present embodiment
It is prompt quick.That is the new weight of label is normalized in the present embodiment.
Continue so that entity A is from tape label as an example, after first iteration, entity A from tape label " health ", " checking card "
The weight of " net is red " is respectively 0.5,0.75 and 0.5.It is 1 that normalization, which is by the maximum value indirect assignment in the new weight of label,
That is, 1 substitution of the label weight 0.75 from tape label " checking card " of entity A is removed remaining 2 weight from tape label respectively
With 0.75 (0.5/0.75=0.6666 ≈ 0.67), 0.67 is obtained, the net as illustrated on the right side of Fig. 3 is red: 0.67, check card 1.0, health
0.67。
Similarly, be normalized rear entity B from tape label " high face value ", " dining room ", " checking card " and " appointment " weight
It is respectively: 0.67,0.67,1.0,0.67.Entity D's divides from the weight after tape label " dining room " and " no peppery not joyous " normalization
It is not: 1.0 and 1.0 (because it is 0.5 maximum value also 0.5 when normalization that two new weights from tape label are equal).
Iterative process of the entity from tape label is described above.It has been observed that one of the present embodiment focuses on label expansion
Exhibition, so-called tag extension is in an iterative process to extend to the label of neighboring entities in the tally set of current entity.In conjunction with
Fig. 3 is explained.
Extension tag refers to the label for existing in the tally set of neighboring entities and being not present in the tally set of current entity.Tool
Body is into Fig. 3, by taking the entity A illustrated on the right side of Fig. 3 as an example, it is known that and entity A absorbs three extension tags of its neighboring entities B,
It is respectively: " high face value ", " dining room " and " appointment ".Since another neighboring entities C of entity A is no label in iteration
, so label of the entity A after a wheel iteration in tally set is only by being extended to the extension mark of entity A from tape label and entity B
Label composition.
Referring to Fig. 3, the new weight calculation of the extension tag of entity A is similar with the aforementioned new weight calculation from tape label, this
In be illustrated by taking a label " high face value " as an example.
For entity A, " high face value " is extension tag, so label (" high face value ") is in first instance (entity A)
On old weight be 0, old weight of the label (" high face value ") on each neighboring entities (i.e. neighboring entities B) is 1.0, then " Gao Yan
New weight W (new)=0+0.5 × sum (1.0 × 0.5)=0.25 of value ".It is equally normalized, by 0.25/0.75 ≈
0.33, i.e., such as the high face value of label of the entity A of the right side signal in Fig. 3: 0.33.
And so on, the extension tag " dining room " that entity A can be obtained and the weight 0.33 and 0.33 after " appointment " normalization.
The entity C that illustrates in Fig. 3 and other entities the difference is that, entity C included mark in first iteration
Label.After a wheel iteration, entity C absorbs the label of its neighboring entities i.e. entity A and entity D, to obtain entity C's
Tally set.It should be noted that the weight of label is also the weight after normalization in entity C, process and previous extension are normalized
The normalized weight of label is similar, that is, the new weight of label (such as health) in first computational entity C determines in new weight most
Big value, is directly set as 1 for maximum value, remaining new weight normalized divided by maximum value after weight, it is therefore, not right here
More explanations are done in the acquisition of the label weight of entity C.
The variable quantity of the new weight of label and weight new after last round of iteration after step (2) calculating epicycle iteration.
With continued reference to Fig. 3, on the basis of the new weight for the label that step (1) determines, variable quantity is calculated.
The calculating process of weight variable quantity are as follows:
Entity A: red 0.33 (1-0.67=0.33) of net checks card 0.0, health 0.33, high face value 0.33, dining room 0.33, about
Meeting 0.33 amounts to 1.65.Note: numerical value here represents the difference of two-wheeled label weight, for example it is logical for netting red difference 0.33
Cross what 1-0.67 was obtained, 1 is the old weight before iteration, and 0.67 is the new weight after iteration.
Entity B: high face value 0.33, dining room 0.33 check card 0.0, appointment 0.33, net red 0.33, and health 0.33 amounts to
1.65。
Entity C: net red 1.0 checks card 1.0, health 1.0, dining room 0.22, without peppery not joyous 0.22, amounts to 3.44.
Entity D: 0 is amounted to.
Variable quantity: (1.65+1.65+3.44+0)/4=1.685.Here 1.65,1.65,3.44,0 be aggregate values, and 4 are
The number for the entity illustrated in Fig. 3.
Step (3) is according to the comparison result of variable quantity and default iteration threshold, it is determined whether stops iteration.
In view of computational efficiency, the number of iterations of the label propagation algorithm of the present embodiment is limited, and the condition of convergence is
Think to restrain if the difference of front and back two-wheeled iteration label weight is within threshold value, and stops iteration.Specifically, by variable quantity
(1.685) it is compared with default iteration threshold (such as 1), stops iteration if variable quantity is less than default iteration threshold.Such as
Fruit variable quantity (1.685) is greater than default iteration threshold, and then return step (one) continues iteration.
So far, it is drawn a portrait according to the label that label is concentrated when stopping iteration for finally constructing the corresponding label of entity.
Furthermore, it is contemplated that mainstream label (or major term) is easier to propagate, it is prominent each in order to avoid the influence of mainstream label
The characteristic of label has personalized and otherness so that the entity of building is drawn a portrait, and carries out comentropy to label in the present embodiment
It calculates, the final weight of label is determined based on comentropy.
Specifically, when obtaining stopping iteration according to abovementioned steps after the tally set of each entity, building entity portrait
Method further include: the comentropy of label is calculated according to probability of occurrence of the label in the tally set of all entities, and based on letter
It ceases the weight of entropy, label on current entity and obtains the final weight of label, to the label in the tally set of each entity according to institute
The size for stating final weight is ranked up, and chooses preceding predetermined number label, and building obtains the label portrait of entity.
Comentropy is calculated as the prior art, is briefly described as follows here for ease of understanding.Comentropy is scaling information amount
Formula, if a things certainty is higher, information content is fewer, if a things is more random, information content is bigger, information
Entropy be for probabilistic measurement means, therefore can based on probability distribution calculate comentropy.
Comentropy calculation formula is expressed as H (X)=- sum, and ((there is the general of some value in the probability × log for some value occur
Rate).
In the present embodiment, the size of information content entrained by measurement label is gone using comentropy.If certain label for
Entity does not have distinction, then it is assumed that the information content of label is few, if high to entity distinction, then it is assumed that label contains much information.
In the present embodiment, list of labels is obtained to tag sorting according to weight size to each entity, list of labels is divided into n sections,
Each section is position section, counts being distributed for different location section of each label in all entities, then should
The comentropy of label is that (certain label appears in the probability × log in certain position section, and (certain label appears in certain position area to W=-sum
Between probability)).It is calculated after comentropy, the result that weight of the label on current entity is multiplied with comentropy
Final weight as label.
The above are the processes of building label portrait.
Label depicting entity is based in the embodiment of the present invention, with the scheme phase for portraying entity based on category etc. in the prior art
Than, the characteristics of entity is portrayed based on label, industry details can be both taken into account, distinguish each entity, and facilitate observation industry overall picture,
For example, after building obtains the label portrait of entity, the label of growth pattern and entity based on entity excavates entity institute
Belong to temperature and the growth point of industry.
Specifically, the method for the present embodiment is after the label portrait of building entity further include: according to the presupposition analysis period
The newly-increased browsing number of users and specified label of interior entity corresponding with specified label determine specified label in weight physically
Growth results；Referred to if growth results are greater than preset threshold according to the size of growth results to specified tag sorting
Calibrate the variable condition information of label.That is, based on label weight physically and in a period of time entity it is newly-increased
Number of users is browsed, the growth results of entity are determined, according to the growth results of each label to tag sorting, to see outgoing label
Variable condition.
For example, the present embodiment mainly excavates new master, uptrend, declining trend these three temperature information.When specific excavation, concern is obtained
Each label newly browse number of users in presupposition analysis period (such as one month), for example, entity " the XX people corresponding with people place label
Increase browsing number of users 100 in one month (such as 2 months) of place " newly, weight of the people place label on XX people place is 0.7, with inn mark
Signing newly-increased browsing number of users of corresponding entity " XX people place " 2 months is 80, and weight of the inn label on " XX people place " is 0.4,
Then the growth results of aforementioned people place label are 70, and the growth results of aforementioned inn label are 32, given threshold 30, it is known that are increased
As a result 70 and growth results 32 be all larger than 30, sort according to the size of growth results, then before people place label comes inn label.
If people place label does not occur in history, people place label is determined as new master.
Preceding description be the absolute increment of label the case where, can also be by the present analysis period of label in practical application
Increment be compared determining sequential growth rate with the increment of a upper analytical cycle, according to sequential growth rate and setting threshold
The comparison result of value determine it is specified label is in the ascendant is in declining trend, to can both take into account industry details or observe
Industry overall picture provides data reference for entity and policymaker.
From the foregoing, it will be observed that construction method and device that the entity of the embodiment of the present application is drawn a portrait, construct real according to the label of extraction
The corresponding label portrait of body facilitates using label as granularity and extracts industrial hot spot and industry growth point, and it is complete to can be not only used for observation industry
Looks, and industry details can be taken into account.Moreover, being different from traditional feature, (for example category, position attribution, sales volume, the value such as evaluation are special
Sign) construction, in the present embodiment using be able to reflect user behavior details, based on similar merchants TGI construction feature, improve spy
The richness and accuracy of expression are levied, so that the entity portrait of building is more variant and comprehensive.
A technical concept is belonged to the construction method of aforementioned entities portrait, the embodiment of the invention also provides a kind of realities
The construction device of body portrait, referring to fig. 4, the construction device 400 of the entity portrait of the present embodiment includes:
Label abstraction module 401 determines the corresponding key of the entity for the text description information according to the entity
Word, according to the feature vector of entity described in browsing behavior information acquisition of the user to the entity, based on the keyword and
Described eigenvector carries out label extraction, obtains label；
Portrait building module 402 is drawn a portrait for constructing the corresponding label of the entity according to the label of extraction.
In one embodiment of the invention, label abstraction module 401 is specifically used for, the feature based on the entity to
Amount calculates the similarity between any two entity；Similarity network is constructed according to the similarity；It is right to obtain first instance institute
Second keyword corresponding to each neighboring entities of the first keyword and the first instance answered, it is crucial by described first
Word and second keyword obtain keyword set；Label extraction is carried out to the keyword in the keyword set, will be taken out
The keyword of taking-up is as label.
In one embodiment of the invention, label abstraction module 401 is specifically used for obtaining each in the keyword set
The keyword abstraction that the TF-IDF score is greater than default score threshold is come out and is used as the mark by the TF-IDF score of keyword
Label.
In one embodiment of the invention, portrait building module 402 is specifically used for initializing in the similarity network
The label of entity obtains the tally set of each entity, includes N number of label in the tally set, N is natural number；Traverse each entity
Tally set is iterated the label in the tally set, constructs the entity according to the tally set of entity each when stopping iteration
Corresponding label portrait；Label in the tally set is iterated and is specifically included: according to power of the label on first instance
Weight, the side right weight of weight and the first instance and the neighboring entities of the label on neighboring entities obtain epicycle
New weight after tag update described in iteration, and the expansion label in the tally set of the neighboring entities is added to described
New tally set is obtained in the tally set of one entity, wherein described to expand in the tally set that label is the neighboring entities not
The label being same as in the tally set of the first instance；The new weight of label changes with last round of after calculating epicycle iteration
The variable quantity of Dai Houxin weight, according to the comparison result of the variable quantity and default iteration threshold, it is determined whether stop iteration, when
Stop the tally set that each entity is determined when iteration.
In one embodiment of the invention, portrait building module 402 is specifically used for each reality after calculating epicycle iteration
Difference of the new weight of the label of body relative to weight new after last round of iteration obtains the difference value and institute
The corresponding aggregate values of entity are stated, average is calculated according to the aggregate values and the number of the entity, obtains the variable quantity；Sentence
Whether the variable quantity that breaks is less than the default iteration threshold, is to stop iteration, otherwise, continues iteration.
In one embodiment of the invention, the construction device 400 of entity portrait further include: drop power module is used for basis
Probability of occurrence of the label in the tally set of all entities calculates the comentropy of the label, and based on the comentropy, described
Weight of the label on current entity obtains the final weight of the label, to the label in the tally set of each entity
It is ranked up according to the size of the final weight, and chooses preceding predetermined number label, building obtains the label of the entity
Portrait.
In one embodiment of the invention, label abstraction module 401 is specifically used for being obtained according to the browsing behavior information
Target group's index TGI value between arbitrarily two entities, the entity is encoded, and by the TGI of the entity
Value is arranged by coded sequence, obtains the feature vector of the entity.
In one embodiment of the invention, label abstraction module 401 is specifically used for the browsing according to first instance
The browsing behavior information of behavioural information and second instance is determined and is also browsed in the user for having browsed the first instance
First number corresponding to the user of the second instance calculates first number and has browsed the user of the first instance
The ratio of number obtains the first accounting；According to corresponding to the user for having browsed the second instance in the standard group of acquisition
Second number calculates the ratio of second number and the number of the standard group, obtains the second accounting；It is accounted for by described first
Than the ratio with second accounting, target group's index TGI value between first instance and second instance is obtained.
In one embodiment of the invention, portrait building module 402 is specifically used for the label that will be extracted and each reality
Keyword in the text description information of body is matched, if matching is consistent, using matched keyword as entity
Be added in sky tally set from tape label, to obtain the tally set of each entity.
In one embodiment of the invention, the construction device 400 of entity portrait further include: tag state excavates module,
For being existed according to the newly-increased browsing number of users and the specified label of entity corresponding with specified label in the presupposition analysis period
The weight physically determines the growth results of the specified label；If the growth results are greater than preset threshold, press
According to the growth results size to the specified tag sorting, obtain the variable condition information of the specified label.
It should be understood that
Algorithm and display be not inherently related to any certain computer, virtual bench or other equipment provided herein.
Various fexible units can also be used together with teachings based herein.As described above, it constructs required by this kind of device
Structure be obvious.In addition, the embodiment of the present application is also not for any particular programming language.It should be understood that can benefit
The content of the embodiment of the present application described herein is realized with various programming languages, and the description done above to language-specific is
In order to disclose the preferred forms of the embodiment of the present application.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that the embodiment of the present application
Embodiment can practice without these specific details.In some instances, well known side is not been shown in detail
Method, structure and technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the disclosure and help to understand one or more of the various inventive aspects,
Above in the description of the exemplary embodiment of the embodiment of the present application, each feature of the embodiment of the present application is grouped together sometimes
Into single embodiment, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention:
The embodiment of the present application i.e. claimed requires features more more than feature expressly recited in each claim.More
Exactly, as reflected in the following claims, inventive aspect is less than single embodiment disclosed above
All features.Therefore, it then follows thus claims of specific embodiment are expressly incorporated in the specific embodiment, wherein
Separate embodiments of each claim as the embodiment of the present application itself.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodiment
Change and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodiment
Member or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use any
Combination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed
All process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint power
Benefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purpose
It replaces.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments mean it is real in the application
It applies within the scope of example and forms different embodiments.For example, in the following claims, implementation claimed
Example it is one of any can in any combination mode come using.
The various component embodiments of the embodiment of the present application can be implemented in hardware, or in one or more processor
The software module of upper operation is realized, or is implemented in a combination thereof.It will be understood by those of skill in the art that can practice
Middle realized using microprocessor or digital signal processor (DSP) is surveyed according to the page performance of the embodiment of the present application embodiment
Trial assembly set in some or all components some or all functions.The embodiment of the present application is also implemented as executing
Some or all device or device programs of method as described herein are (for example, computer program and computer journey
Sequence product).Such program for realizing the embodiment of the present application can store on a computer-readable medium, or can have one
The form of a or multiple signals.Such signal can be downloaded from an internet website to obtain, or mention on the carrier signal
For, or be provided in any other form.
For example, Fig. 5 is the structural schematic diagram of the electronic equipment in the application one embodiment.The electronic equipment 500 includes:
Processor 510, and it is stored with the memory 520 for the computer program that can be run on the processor 510.Processor 510,
For executing each step of method in the embodiment of the present application when executing the computer program in the memory 520.Memory
520 can be the electricity of such as flash memory, EEPROM (electrically erasable programmable read-only memory), EPROM, hard disk or ROM etc
Sub memory.Memory 520 has depositing for computer program 531 of the storage for executing any method and step in the above method
Store up space 530.Computer program 531 can read or be written to this from one or more computer program product
Or in multiple computer program products.These computer program products include such as hard disk, compact-disc (CD), storage card or
The program code carrier of floppy disk etc.Such computer program product is usually computer-readable storage described in such as Fig. 6
Medium.
Fig. 6 is the structural schematic diagram of one of the application one embodiment computer readable storage medium.The computer
Readable storage medium storing program for executing 600 is stored with for executing the computer program 531 according to the method and step of the embodiment of the present application, can be by
The processor 510 of electronic equipment 500 is read, and when computer program 531 is run by electronic equipment 500, leads to the electronic equipment
500 execute each step in method described above, specifically, the calculating journey of the computer-readable recording medium storage
Sequence 531 can execute method shown in any of the above-described embodiment.Computer program 531 can be compressed in a suitable form.
The embodiment of the present application is carried out it should be noted that above-described embodiment illustrates rather than the embodiment of the present application
Limitation, and those skilled in the art can be designed alternative embodiment without departing from the scope of the appended claims.
In the claims, any reference symbol between parentheses should not be configured to limitations on claims.Word " packet
Containing " do not exclude the presence of element or step not listed in the claims.Word "a" or "an" located in front of the element is not arranged
Except there are multiple such elements.The embodiment of the present application can by means of include several different elements hardware and by means of
Properly programmed computer is realized.In the unit claims listing several devices, several in these devices can
To be to be embodied by the same item of hardware.The use of word first, second, and third does not indicate any sequence.
These words can be construed to title.
Claims (14)
1. a kind of construction method of entity portrait characterized by comprising
The corresponding keyword of the entity is determined according to the text description information of the entity；
According to the feature vector of entity described in browsing behavior information acquisition of the user to the entity；
Label extraction is carried out based on the keyword and described eigenvector, obtains label；
The corresponding label portrait of the entity is constructed according to the label of extraction.
2. the method as described in claim 1, which is characterized in that described to be carried out based on the keyword and described eigenvector
Label extracts, and obtaining label includes:
Based on the feature vector of the entity, the similarity between any two entity is calculated；
Similarity network is constructed according to the similarity；
It is real to obtain each neighbours of the first keyword corresponding to first instance and the first instance in the similarity network
Second keyword corresponding to body obtains keyword set by first keyword and second keyword；
Label extraction is carried out to the keyword in the keyword set, using the keyword extracted as label.
3. method according to claim 2, which is characterized in that the keyword in the keyword set carries out label
Extraction includes:
The TF-IDF score is greater than default score threshold by the TF-IDF score for obtaining each keyword in the keyword set
Keyword abstraction come out be used as the label.
4. method according to claim 2, which is characterized in that described corresponding according to the label of the extraction building entity
Label portrait include:
The label for initializing entity in the similarity network obtains the tally set of each entity, includes N number of mark in the tally set
Label, N is natural number；
The tally set for traversing each entity is iterated the label in the tally set, the mark of each entity when according to stopping iteration
Label collection constructs the corresponding label portrait of the entity；
Label in the tally set is iterated and is specifically included:
According to weight of the label on first instance, weight and the first instance and institute of the label on neighboring entities
State the side right weight of neighboring entities, the new weight after obtaining tag update described in epicycle iteration, and by the mark of the neighboring entities
The expansion label that label are concentrated, which is added in the tally set of the first instance, obtains new tally set,
Wherein, described expand in the tally set that label is the neighboring entities is different from the tally set of the first instance
Label；
The variable quantity of the new weight of label and weight new after last round of iteration after calculating epicycle iteration, according to the variable quantity
With the comparison result of default iteration threshold, it is determined whether stop iteration, the tally set of each entity is determined when stopping iteration.
5. method as claimed in claim 4, which is characterized in that it is described calculate epicycle iteration after label the new weight with it is upper
The variable quantity of new weight after one wheel iteration, according to the comparison result of the variable quantity and default iteration threshold, it is determined whether stop
Iteration includes:
The new weight of the label of each entity is relative to weight new after last round of iteration after calculating epicycle iteration
The difference value is obtained aggregate values corresponding with the entity, according to the number of the aggregate values and the entity by difference
Average is calculated, the variable quantity is obtained；
Judge whether the variable quantity is less than the default iteration threshold, be to stop iteration, otherwise, continues iteration.
6. method as claimed in claim 4, which is characterized in that the tally set of each entity when obtaining stopping iteration
Later, this method further include:
The comentropy of the label is calculated according to probability of occurrence of the label in the tally set of all entities, and is based on the information
The weight of entropy, the label on current entity obtains the final weight of the label,
Label in the tally set of each entity is ranked up according to the size of the final weight, and is chosen preceding pre-
If a several labels, building obtains the label portrait of the entity.
7. the method as described in claim 1, which is characterized in that described to be obtained according to browsing behavior information of the user to the entity
The feature vector of the entity includes:
According to the browsing behavior information, target group's index TGI value between any two entity is obtained, the entity is carried out
Coding, and the TGI value of the entity is arranged by coded sequence, obtain the feature vector of the entity.
8. the method for claim 7, which is characterized in that it is described according to the browsing behavior information, it is real to obtain any two
Target group's index TGI value between body includes:
According to the browsing behavior information of first instance and the browsing behavior information of second instance, determination has browsed institute
It states and has also browsed the first number corresponding to the user of the second instance in the user of first instance, calculate first number
With the ratio for the number of users for having browsed the first instance, the first accounting is obtained；
According to the second number corresponding to the user for having browsed the second instance in the standard group of acquisition, described second is calculated
The ratio of number and the number of the standard group, obtains the second accounting；
By the ratio of first accounting and second accounting, the target group obtained between first instance and second instance refer to
Number TGI value.
9. method as claimed in claim 4, which is characterized in that the label of entity in the initialization similarity network,
The tally set for obtaining each entity includes:
The label of extraction is matched with the keyword in the text description information of each entity,
It is each to obtain using matched keyword being added in sky tally set from tape label as entity if matching is consistent
The tally set of entity.
10. method as claimed in any one of claims 1-9 wherein, which is characterized in that draw a portrait it in the label for constructing the entity
Afterwards, this method further include:
Existed according to the newly-increased browsing number of users of entity corresponding with specified label in the presupposition analysis period and the specified label
The weight physically determines the growth results of the specified label；
If the growth results are greater than preset threshold, according to the size of the growth results to the specified tag sorting,
Obtain the variable condition information of the specified label.
11. a kind of construction device of entity portrait characterized by comprising
Label abstraction module determines the corresponding keyword of the entity for the text description information according to the entity, according to
The feature vector of entity described in browsing behavior information acquisition of the user to the entity is based on the keyword and the feature
Vector carries out label extraction, obtains label；
Portrait building module is drawn a portrait for constructing the corresponding label of the entity according to the label of extraction.
12. device as claimed in claim 11, which is characterized in that the label abstraction module is specifically used for, and is based on the reality
The feature vector of body calculates the similarity between any two entity；Similarity network is constructed according to the similarity；Obtain the
Second keyword corresponding to each neighboring entities of first keyword and the first instance corresponding to one entity, by institute
It states the first keyword and second keyword obtains keyword set；Label is carried out to the keyword in the keyword set
It extracts, using the keyword extracted as label.
13. a kind of electronic equipment, which is characterized in that the electronic equipment includes: processor, and being stored with can be on a processor
The memory of the computer program of operation；
Wherein, the processor, for appointing in perform claim requirement 1-10 when executing the computer program in the memory
Method described in one.
14. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program quilt
Processor realizes method of any of claims 1-10 when executing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910308951.8A CN110134794B (en) | 2019-04-17 | 2019-04-17 | Method and device for constructing entity portrait |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910308951.8A CN110134794B (en) | 2019-04-17 | 2019-04-17 | Method and device for constructing entity portrait |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110134794A true CN110134794A (en) | 2019-08-16 |
| CN110134794B CN110134794B (en) | 2020-08-14 |
Family
ID=67570333
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910308951.8A Active CN110134794B (en) | 2019-04-17 | 2019-04-17 | Method and device for constructing entity portrait |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110134794B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079026A (en) * | 2019-11-28 | 2020-04-28 | 精硕科技（北京）股份有限公司 | Method, storage medium and device for determining character impression data |
| CN111597434A (en) * | 2020-04-10 | 2020-08-28 | 广东机电职业技术学院 | Takeaway recommendation method, system, device and medium based on user portrait |
| CN112052365A (en) * | 2020-09-03 | 2020-12-08 | 中国银行股份有限公司 | Cross-border scene portrait construction method and device |
| CN117688250A (en) * | 2024-02-04 | 2024-03-12 | 国网湖北省电力有限公司信息通信公司 | Unified data dynamic service management system and method suitable for electric power full scene |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760541A (en) * | 2016-03-15 | 2016-07-13 | 传成文化传媒（上海）有限公司 | Video food ordering recommendation method and system |
| CN106682013A (en) * | 2015-11-09 | 2017-05-17 | 阿里巴巴集团控股有限公司 | Method and device used for data pushing |
| CN107944946A (en) * | 2017-11-03 | 2018-04-20 | 清华大学 | Commercial goods labels generation method and device |
| US10120880B2 (en) * | 2014-06-26 | 2018-11-06 | Amazon Technologies, Inc. | Automatic image-based recommendations using a color palette |
-
2019
- 2019-04-17 CN CN201910308951.8A patent/CN110134794B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10120880B2 (en) * | 2014-06-26 | 2018-11-06 | Amazon Technologies, Inc. | Automatic image-based recommendations using a color palette |
| CN106682013A (en) * | 2015-11-09 | 2017-05-17 | 阿里巴巴集团控股有限公司 | Method and device used for data pushing |
| CN105760541A (en) * | 2016-03-15 | 2016-07-13 | 传成文化传媒（上海）有限公司 | Video food ordering recommendation method and system |
| CN107944946A (en) * | 2017-11-03 | 2018-04-20 | 清华大学 | Commercial goods labels generation method and device |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079026A (en) * | 2019-11-28 | 2020-04-28 | 精硕科技（北京）股份有限公司 | Method, storage medium and device for determining character impression data |
| CN111079026B (en) * | 2019-11-28 | 2023-11-24 | 北京秒针人工智能科技有限公司 | Method, storage medium and device for determining character impression data |
| CN111597434A (en) * | 2020-04-10 | 2020-08-28 | 广东机电职业技术学院 | Takeaway recommendation method, system, device and medium based on user portrait |
| CN111597434B (en) * | 2020-04-10 | 2023-04-18 | 广东机电职业技术学院 | Takeout recommendation method, system, device and medium based on user portrait |
| CN112052365A (en) * | 2020-09-03 | 2020-12-08 | 中国银行股份有限公司 | Cross-border scene portrait construction method and device |
| CN112052365B (en) * | 2020-09-03 | 2024-05-24 | 中国银行股份有限公司 | Cross-border scene portrait construction method and device |
| CN117688250A (en) * | 2024-02-04 | 2024-03-12 | 国网湖北省电力有限公司信息通信公司 | Unified data dynamic service management system and method suitable for electric power full scene |
| CN117688250B (en) * | 2024-02-04 | 2024-04-16 | 国网湖北省电力有限公司信息通信公司 | Unified data dynamic service management system and method suitable for electric power full scene |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110134794B (en) | 2020-08-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110134794A (en) | A kind of construction method and device of entity portrait | |
| Bai et al. | Design reuse oriented partial retrieval of CAD models | |
| CN109299362B (en) | Similar enterprise recommendation method and device, computer equipment and storage medium | |
| CN103164804B (en) | The information-pushing method of a kind of personalization and device | |
| He et al. | Heterotic models from vector bundles on toric Calabi-Yau manifolds | |
| CN109087135A (en) | The method for digging and device, computer equipment and readable medium that user is intended to | |
| CN109919719A (en) | Information-pushing method calculates equipment and computer storage medium | |
| CN106598949B (en) | A kind of determination method and device of word to text contribution degree | |
| CN103279486B (en) | It is a kind of that the method and apparatus of relevant search are provided | |
| CN106096028A (en) | Historical relic indexing means based on image recognition and device | |
| Mitchell et al. | Non-manifold level sets: A multivalued implicit surface representation with applications to self-collision processing | |
| CN107315827A (en) | The method and its device of a kind of correlation recommendation in electronic reading | |
| CN106484698A (en) | A kind of method for pushing of search keyword and device | |
| CN104142945B (en) | A kind of search method and device based on query word | |
| Kwon et al. | User-assisted integrated method for controlling level of detail of large-scale B-rep assembly models | |
| CN103425680A (en) | Selection method and system for page advertisement demonstration | |
| WO2014205046A1 (en) | System and method for text mining documents | |
| CN105468628A (en) | Sorting method and apparatus | |
| CN104915440A (en) | Commodity de-duplication method and system | |
| Jiang et al. | An efficient community detection algorithm using greedy surprise maximization | |
| CN115423555A (en) | Commodity recommendation method and device, electronic equipment and storage medium | |
| CN106407316A (en) | Topic model-based software question and answer recommendation method and device | |
| CN110389963A (en) | The recognition methods of channel effect, device, equipment and storage medium based on big data | |
| CN109522275A (en) | Label method for digging, electronic equipment and the storage medium of content are produced based on user | |
| US20110099183A1 (en) | Search-based filtering for property grids |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |