CN109460508B - Efficient spam comment user group detection method - Google Patents
Efficient spam comment user group detection method Download PDFInfo
- Publication number
- CN109460508B CN109460508B CN201811177783.5A CN201811177783A CN109460508B CN 109460508 B CN109460508 B CN 109460508B CN 201811177783 A CN201811177783 A CN 201811177783A CN 109460508 B CN109460508 B CN 109460508B
- Authority
- CN
- China
- Prior art keywords
- comment
- group
- spam
- comments
- user
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 16
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000000034 method Methods 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 19
- 230000006399 behavior Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000001680 brushing effect Effects 0.000 claims description 3
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims description 2
- 230000017105 transposition Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 description 6
- 238000012804 iterative process Methods 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 5
- 230000001133 acceleration Effects 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007430 reference method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an efficient spam comment user group detection method which is characterized in that candidate groups (each group is required to be composed of at least 2 persons and comment 3 products together) are obtained based on comment data of products on an E-commerce website, the basic spam information characteristics of each product node, user nodes and group nodes in a heterogeneous network based on self and the spam influence characteristics based on relationship are extracted, the influence of the basic spam information of each node and other two types of nodes on the spam information of each node is considered, the spam information of each node is obtained, the spam information of the group nodes is included, and the groups higher than a certain threshold value are identified as spam comment user groups. Meanwhile, the optimized GroupRank algorithm is adopted, so that the accuracy rate and the performance are higher.
Description
Technical Field
The invention belongs to the technical field of data mining, and particularly relates to an efficient spam comment user group detection method.
Background
With the coming of the mobile internet era and the improvement of the internet of things, online shopping gradually becomes a novel important consumption mode, more and more contents generated by multiple users appear in network application, and most consumers can publish their shopping experiences and use thoughts and opinions on products on an e-commerce platform after shopping. On one hand, according to the survey report of the united states Cone 2011, 64% of users refer to existing user comments before purchasing behaviors, and the comments made by the users have certain influence on the consumption behaviors of potential consumers and have commercial value; on the other hand, the comments of the users contain a great deal of information such as evaluation and preference degree of the consumers on various aspects of the products, and the information can help enterprises to know the preference and consumption requirements of the consumers more conveniently, find problems such as product quality and know the advantages and the defects of the product performance of the enterprises. For the E-commerce service platform, the characteristics of the commodities most concerned by consumers are known through the commodity comment information, the consumers are guided to evaluate the commodities more comprehensively in the commodity comment system, and the comment information quality and the website public praise are improved.
The comments of the e-commerce website are derived from the real experience of a user on a specific product or service, the purchasing decision of a future customer is directly influenced, and the positive idea can bring huge economic benefit and wealth for organizations and individuals. Because consumers of the same kind of commodities tend to visit stores with large sales volume, large number of comments and large number of good comments, but under the drive of interests, merchants often hire water army to do false transactions and then swipe the number of comments or do activities such as full N-word good comment return, and others can issue false comments to try to make unfair comments on some products, such as writing good comments to promote the products, or deliberately making poor comments to damage the reputation of a certain product, so that the consumers are misled, and the false comments are called spam comments.
Unlike other types of spam (e.g., spam email), spam comments are very difficult to find, mainly because spam comment users can easily disguise themselves, resulting in a bottleneck in algorithms that rely on underlying textual language features and behavioral characteristics to detect spam comments, which is difficult to identify spam comment and spam comment users. At present, research on the field of spam comments is basically developed around detection of spam comments and spam comment users, however, a spam comment user group has stronger destructiveness, and due to the fact that many members of a group write false comments, the comment emotion of a product can be completely controlled. And it is found that compared with the detection difficulty of the spam comment and the spam comment users, the spam comment user group is easier to detect, so that the spam comment user group is more significant to detect.
Since the comment user behavior proposed in the existing spam comment methods is not sufficient to capture a spam comment user group, it is necessary to find a more complex and complementary framework.
Disclosure of Invention
In view of the above, the invention provides an efficient spam comment user group detection method, which includes capturing the mutual relation among comment user groups, comment users and products, mining the product comment data of an e-commerce platform by using a heterogeneous network iterative algorithm GroupRank, obtaining spam information of the groups by considering the influence of product nodes and user nodes on the group nodes and the basic spam information of the groups, and identifying the spam comment user groups.
An efficient spam comment user group detection method comprises the following steps:
(1) preprocessing comment data of the E-commerce product;
(2) extracting basic features of groups, products and users based on the preprocessed comment data and integrating the basic features into a vector form, wherein the group is composed of at least 2 users, and the users comment 3 products together;
(3) extracting three groups of corresponding relation characteristics among the group, the product and the user and integrating the three groups of relation characteristics into a matrix form;
(4) calculating the junk information scores of all groups through a GroupRank algorithm according to the basic characteristics and the relational characteristics;
(5) setting a proper threshold value, and classifying all groups into a garbage group and a non-garbage group through comparison between the garbage information score and the threshold value.
Further, the preprocessing of the comment data in the step (1) comprises filtering low-frequency users, low-quality comments and low-sales commodities; for low-quality comments, the comments with too low information quantity, namely below a set threshold value, are screened out through the comment length and the richness, and then spam comments and advertisement comments are removed according to manual judgment and machine learning; for low-frequency users, namely the users can not accurately identify the effective comments which are less than the set threshold value, the comments issued by the users are removed; for a low-sales commodity, namely if the comment of a commodity is very few, namely lower than a set threshold value, the commodity is considered to have basically no comment brushing behavior, and all comments of the commodity are rejected.
Further, the basic features of the group, the product and the user are extracted in the step (2), wherein the basic features of the group are the similarity of the behavior information of the group members on all the products commented together (the higher the similarity is, the more likely the group commented on the products according to the task is), and the behavior information comprises comment texts, scores, time windows and comment positions; the basic characteristics of the users are the consistency of the overall comments of the users (spam comment users generally brush the comments deliberately, so that the comment consistency is higher), and the consistency is expressed in the aspects of user rating, comment time and comment position; the basic characteristic of the product is the deviation degree of the comments, and since the product can have the comments which are intentionally commented or badly commented, the comments deviating from the real condition generate the deviation feeling in the overall comments, which includes the deviation in the scores and the comment time.
Further, in the step (3), in order to quantify the magnitude of influence of spam among the three types of entities, namely, group, product and user, appropriate relationship features conforming to each group of relationships are extracted from the three groups of relationships, namely, group-product, user-product and group-user.
Further, the GroupRank algorithm in step (4) is as follows:
wherein: s1～S5Are all intermediate variables, BU、BP、BGBasic feature vectors, A, for users, products, groups, respectivelyPG、AUP、AGUThe relation feature matrixes are respectively group-product, user-product and group-user;
and
group spam vectors in the t-1 th iteration process and the t-th iteration process respectively comprise spam scores of all groups; alpha is a set weight coefficient,Trepresenting the transpose of the matrix, t being a natural number greater than 0, | | | | | | non-woven phosphor1Is a norm of 1; initialization
When in use
Algorithm convergence and output
δ is a set convergence threshold, | | | | | non-calculation∞Is an infinite norm.
Preferably, since a large number of matrix operations are used in the iterative process, which consumes very much computing resources, S is respectively pre-aligned1～S5And
carrying out standardization, and obtaining the standard by converting 6 groups of formulas in the GroupRank algorithm
W is a transformation matrix obtained by transformation integration.
The method comprises the steps of obtaining candidate groups (each group is required to be at least composed of 2 persons and at least comments 3 products together) based on comment data of products on an E-commerce website, extracting basic spam information characteristics and relation-based spam influence characteristics of each product node, user nodes and group nodes in a heterogeneous network, obtaining spam information of each node by considering the influence of the basic spam information of each node and other two types of nodes on the spam information of each node, including the spam information of the group nodes, and identifying the group higher than a certain threshold value as a spam comment user group; therefore, the invention has the following beneficial technical effects:
(1) based on the idea that webpage nodes in the PageRank transmit webpage authority values in a network structure, the invention provides a heterogeneous network model consisting of three types of nodes to capture a more complex relationship among comment user groups, comment users and products, and more comprehensively analyzes various possible factors than other spam comment user group identification algorithms.
(2) The invention carries out iterative computation based on a heterogeneous network model, can deduce that the junk information of one node influences the junk information of another node through the influence of the junk according to the relation among the 3 types of nodes, simultaneously considers the basic junk information of the nodes and uses an adjusting factor to weight the basic junk information so as to adjust the proportion of the basic junk information to the influence of the junk based on the relation.
(3) The optimized GroupRank algorithm is adopted, an experiment is carried out on 120 ten thousand comment data sets of products on the E-commerce website of Amazon China, and the experiment is compared with other identification methods commonly used in the field, and the experiment shows that the algorithm can achieve higher accuracy rate and performance.
Drawings
FIG. 1 is a schematic overall flow chart of the method of the present invention.
Fig. 2 is a flow chart of the GroupRank algorithm.
FIG. 3 is a schematic diagram of accelerated optimization of the GroupRank algorithm.
Detailed Description
In order to more specifically describe the present invention, the following detailed description is provided for the technical solution of the present invention with reference to the accompanying drawings and the specific embodiments.
The spam comment user group detection method disclosed by the invention is mainly divided into six parts as shown in FIG. 1: comment reading, data preprocessing and analysis, basic feature extraction, relation feature extraction, GroupRank iterative computation and algorithm acceleration, and spam group identification, wherein:
the comment reading part mainly defines an interface for data input and output, and data can be acquired or output from different channels and types through duplicating an interface function at any time.
Data preprocessing and analysis: the targeted data is comment data of E-commerce products, the data is relatively complex and not standard, and certain cleaning is needed to improve the detection accuracy, so comment contents, users and groups are mainly screened in the data preprocessing part, low-frequency users, low-quality comments and low-sales commodities are filtered, and main data are obtained. The method mainly comprises the following four steps: the method comprises the following steps that firstly, for low-quality comments, comments with low obvious information content are screened according to the comment length and the comment richness, and then spam comments and advertisement comments are removed according to simple manual comment screening and machine learning; secondly, screening low-frequency users, wherein the users can not accurately identify the users who give too few effective comments; the third step is directed at the commodity, if the comment of one commodity is few, the commodity is considered to have no comment brushing behavior basically; and finally, mining candidate high-similarity user groups based on a machine learning method of frequent mining according to the screened data.
Extracting basic characteristics: the basic characteristics of the group, the product and the user are based on the overall characteristics of the group, the user and the product, the overall similarity, consistency or deviation degree is mainly extracted to serve as the basic junk information characteristics, and the indexes reflect the junk information of the node to a certain degree. The core of the group basic characteristics is the similarity of the behaviors of the group on all the products which are commented together, the higher the similarity is, the more likely the group is to comment the products spam according to tasks, and the behaviors comprise comment texts, scores, time windows, comment positions and the like; the core of the basic characteristics of the comment users is the consistency of the overall comments of the users, and the spam comment users generally have the meaning of good comment, so that the comment consistency is higher. The consistency of the behaviors is realized in the aspects of user score, user comment time and user comment position; the core of the basic characteristics of the product is the deviation degree of the comments, because the product has the comments which are intentionally commented or badly commented, the comments deviate from the comments of the real users, and the deviation sense is generated in the overall comments, including the deviation in the aspects of the product rating, the product comment time and the like.
Extracting the relation features: because the influence degrees among the entities are different, and the spam information degrees transmitted among the entities are also different, the relationship characteristics are extracted from different entity relationships to represent different spam information influence. In order to quantify the magnitude of the influence of the garbage among the entities, the relationship feature extraction part extracts proper relationship features which accord with each relationship from 3 relationships of a comment user group-product, a comment user-product and a comment user group-comment user; for example, the suspicious degrees of behaviors of the same group on different products are different, the spam degrees of comments on different products by the same comment user are different, and the larger the strength of the influence is, the larger the spam influence between the comment user and the comment user is.
GroupRank iterative computation and algorithm acceleration: after the junk information is spread between the nodes, the junk information of each node can be pushed out by the junk information of other two types of nodes, so that the junk information of the group is initialized to be basic junk information, an iterative process is started until the last update, and S of the previous and subsequent two times is completedGIs less than a threshold delta, at which point the iterative process converges to a stable SGThe spam scores of all groups are obtained. Because a large amount of matrix operation is used in the iteration process, the calculation resources are consumed very much, and the acceleration effect is achieved through the advanced iterative calculation of the matrix W.
The process of the GroupRank algorithm is as shown in fig. 2, at the beginning of the algorithm, firstly, initializing the spam of all groups as their basic characteristic values; the iterative process is then started, in a first step from SGDeducing product spam information S from the current valueP(ii) a In the second step, useNew SPIn deducing the user spam SU(ii) a In the third step using SUAnd deducing the group garbage information, thus completing a forward loop deduction. Starting from the fourth step, the reverse loop derivation is first performed by SGDerivation of user spam SUThen from the updated SUDeduct the junk information S of the productPFinally, in the sixth step, group garbage information S is derivedG. Thus, a round of iteration process is completed, the spam degree is standardized in each step process, and the iteration group spam degree is continuously updated until the last update leads to the two previous and next group spam information SGIs less than a threshold delta, at which point the iterative process converges to a stable SG，SGThe spam score of all groups is included, and the actual spam score of the group can be solved through the iterative process.
The algorithm accelerating part achieves the accelerating effect through the advanced iterative calculation of the matrix W, and as shown in figure 3, the S is respectively pre-calculatedP、SU、SGThe standardization is carried out, then 6 formulas which are derived from each other among three types of nodes are transformed, and the calculation process of the GroupRank can be simply summarized as follows:
let W be ZPGZUPZGUZUGZPUZGPThen the calculation process can be converted into:
an iteration process at this time is equivalent to the calculation process plus a normalization and whether the normalization is converged; the GroupRank needs to perform W and S in each iterationGThe matrix multiplication operation between them, whose time complexity can be considered as O (t | E | may,where | E | is the number of non-zero elements in A and t is the total number of iterations. In the process of generating the matrix W, because the matrix W only needs a few times of matrix calculation, compared with iterative calculation, the matrix W can be ignored; the actual computation is rather fast because the relationship matrix is very sparse and follows the distribution of comments, resulting in a large number of null computations in the matrix computation. In addition, the GroupRank is an algorithm capable of performing power iteration, and the effect of acceleration is achieved by performing advanced iterative computation on the matrix W, so that the optimized GroupRank is an efficient spam comment user group identification method.
Identifying a garbage group: and setting a proper threshold value, and dividing the candidate group into a garbage group and a non-garbage group. The part is used for identifying the candidate groups higher than a certain spam threshold value gamma as spam comment user groups, and in order to identify the spam comment user groups more accurately, the ROC can be manually marked and calculated under different scenes, so that a better threshold value can be selected, and a better classification effect can be achieved.
According to the GroupRank algorithm, a classification experiment is carried out on comment data of 120 ten thousand products captured from Amazon China, comparison is carried out with other garbage group identification methods, and the F-score value of GroupRank (when gamma is 0.67) is found to be the highest and reaches 0.917; and the F-score of the other recognition algorithms SVM, KNN, GC are 0.834, 0.908, 0.916, respectively. Therefore, experiments prove that the GroupRank provided by the invention has high identification accuracy rate in identifying the spam comment user group, is really superior to other reference methods, and the optimized GroupRank algorithm is also an efficient spam comment user group identification method.
The embodiments described above are presented to enable a person having ordinary skill in the art to make and use the invention. It will be readily apparent to those skilled in the art that various modifications to the above-described embodiments may be made, and the generic principles defined herein may be applied to other embodiments without the use of inventive faculty. Therefore, the present invention is not limited to the above embodiments, and those skilled in the art should make improvements and modifications to the present invention based on the disclosure of the present invention within the protection scope of the present invention.
Claims (6)
1. An efficient spam comment user group detection method comprises the following steps:
(1) preprocessing comment data of the E-commerce product;
(2) extracting basic features of groups, products and users based on the preprocessed comment data and integrating the basic features into a vector form, wherein the group is composed of at least 2 users, and the users comment 3 products together;
(3) extracting three groups of corresponding relation characteristics among the group, the product and the user and integrating the three groups of relation characteristics into a matrix form;
(4) calculating the junk information scores of all groups through a GroupRank algorithm according to the basic characteristics and the relational characteristics;
(5) setting a proper threshold value, and classifying all groups into a garbage group and a non-garbage group through comparison between the garbage information score and the threshold value.
2. The spam comment user group detection method of claim 1 wherein: preprocessing the comment data in the step (1) comprises filtering low-frequency users, low-quality comments and low-sales commodities; for low-quality comments, the comments with too low information quantity, namely below a set threshold value, are screened out through the comment length and the richness, and then spam comments and advertisement comments are removed according to manual judgment and machine learning; for low-frequency users, namely the users can not accurately identify the effective comments which are less than the set threshold value, the comments issued by the users are removed; for a low-sales commodity, namely if the comments of a commodity are few, namely lower than a set threshold value, the commodity is considered to have no comment brushing behavior, and all the comments of the commodity are rejected.
3. The spam comment user group detection method of claim 1 wherein: extracting basic features of the group, the product and the user in the step (2), wherein the basic features of the group are the similarity of behavior information of the group members on all products commented together, and the behavior information comprises comment texts, scores, time windows and comment positions; the basic characteristics of the users are the consistency of the overall comments of the users, and the consistency is expressed in the aspects of user scores, comment time and comment positions; the basic characteristic of the product is the deviation degree of the comments, and since the product can have the comments which are intentionally commented or badly commented, the comments deviating from the real condition generate the deviation feeling in the overall comments, which includes the deviation in the scores and the comment time.
4. The spam comment user group detection method of claim 1 wherein: and (3) extracting the relation characteristics which accord with each group of relations and are proper from the three groups of relations of group-product, user-product and group-user in order to quantify the influence of the garbage among the three types of entities of group, product and user.
5. The spam comment user group detection method of claim 1 wherein: the GroupRank algorithm in step (4) is as follows:
wherein: s1～S5Are all intermediate variables, BU、BP、BGBasic feature vectors, A, for users, products, groups, respectivelyPG、AUP、AGUGroup-product, user-product, group-user associations, respectivelyA family feature matrix;
and
group spam vectors in the t-1 th iteration process and the t-th iteration process respectively comprise spam scores of all groups; alpha is a set weight coefficient, T represents matrix transposition, T is a natural number greater than 0, | | | | | non-woven phosphor1Is a norm of 1; initialization
When in use
Algorithm convergence and output
δ is a set convergence threshold, | | | | | non-calculation∞Is an infinite norm.
6. The spam comment user group detection method of claim 5 wherein: respectively pre-align with S1～S5And
carrying out standardization, and obtaining the standard by converting 6 groups of formulas in the GroupRank algorithm
W is a transformation matrix obtained by transformation integration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811177783.5A CN109460508B (en) | 2018-10-10 | 2018-10-10 | Efficient spam comment user group detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811177783.5A CN109460508B (en) | 2018-10-10 | 2018-10-10 | Efficient spam comment user group detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109460508A CN109460508A (en) | 2019-03-12 |
| CN109460508B true CN109460508B (en) | 2021-10-15 |
Family
ID=65607459
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811177783.5A Active CN109460508B (en) | 2018-10-10 | 2018-10-10 | Efficient spam comment user group detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460508B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110442798B (en) * | 2019-07-03 | 2021-10-08 | 华中科技大学 | Spam comment user group detection method based on network representation learning |
| CN112308352A (en) * | 2019-07-30 | 2021-02-02 | 阿里巴巴集团控股有限公司 | Order identification method and device and electronic equipment |
| CN111259140B (en) * | 2020-01-13 | 2023-07-28 | 长沙理工大学 | False comment detection method based on LSTM multi-entity feature fusion |
| CN111275526B (en) * | 2020-01-20 | 2021-04-13 | 南京财经大学 | E-commerce water force identification method based on range difference |
| CN111242647B (en) * | 2020-01-20 | 2021-04-13 | 南京财经大学 | Method for identifying malicious user based on E-commerce comment |
| CN111310064B (en) * | 2020-02-18 | 2023-05-05 | 重庆锐云科技有限公司 | Computing method, device and storage medium for eliminating business comment |
| CN113627960A (en) * | 2020-05-06 | 2021-11-09 | 山东科技大学 | Water army group detection method and device |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101393555A (en) * | 2008-09-09 | 2009-03-25 | 浙江大学 | Rubbish blog detecting method |
| CN101593204A (en) * | 2009-06-05 | 2009-12-02 | 北京大学 | A kind of emotion tendency analysis system based on news comment webpage |
| US8612436B1 (en) * | 2011-09-27 | 2013-12-17 | Google Inc. | Reverse engineering circumvention of spam detection algorithms |
| CN103514174A (en) * | 2012-06-18 | 2014-01-15 | 北京百度网讯科技有限公司 | Text categorization method and device |
| CN106056455A (en) * | 2016-06-02 | 2016-10-26 | 南京邮电大学 | Group and place recommendation method based on location and social relationship |
| CN107153642A (en) * | 2017-05-16 | 2017-09-12 | 华北电力大学 | A kind of analysis method based on neural network recognization text comments Sentiment orientation |
| CN107977798A (en) * | 2017-12-21 | 2018-05-01 | 中国计量大学 | A kind of risk evaluating method of e-commerce product quality |
| CN108230085A (en) * | 2017-11-27 | 2018-06-29 | 重庆邮电大学 | A kind of commodity evaluation system and method based on user comment |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8010482B2 (en) * | 2008-03-03 | 2011-08-30 | Microsoft Corporation | Locally computable spam detection features and robust pagerank |
-
2018
- 2018-10-10 CN CN201811177783.5A patent/CN109460508B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101393555A (en) * | 2008-09-09 | 2009-03-25 | 浙江大学 | Rubbish blog detecting method |
| CN101593204A (en) * | 2009-06-05 | 2009-12-02 | 北京大学 | A kind of emotion tendency analysis system based on news comment webpage |
| US8612436B1 (en) * | 2011-09-27 | 2013-12-17 | Google Inc. | Reverse engineering circumvention of spam detection algorithms |
| CN103514174A (en) * | 2012-06-18 | 2014-01-15 | 北京百度网讯科技有限公司 | Text categorization method and device |
| CN106056455A (en) * | 2016-06-02 | 2016-10-26 | 南京邮电大学 | Group and place recommendation method based on location and social relationship |
| CN107153642A (en) * | 2017-05-16 | 2017-09-12 | 华北电力大学 | A kind of analysis method based on neural network recognization text comments Sentiment orientation |
| CN108230085A (en) * | 2017-11-27 | 2018-06-29 | 重庆邮电大学 | A kind of commodity evaluation system and method based on user comment |
| CN107977798A (en) * | 2017-12-21 | 2018-05-01 | 中国计量大学 | A kind of risk evaluating method of e-commerce product quality |
Non-Patent Citations (2)
| Title |
|---|
| "PageRank optimization applied to spam detection,";O. Fercoq,;《2012 6th International Conference on Network Games, Control and Optimization》;20130325;第127-134页 * |
| "基于评论者关系的垃圾评论者识别研究";徐小婷 等;《集美大学学报(自然科学版)》;20160331;第146-152页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109460508A (en) | 2019-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460508B (en) | Efficient spam comment user group detection method | |
| US11995702B2 (en) | Item recommendations using convolutions on weighted graphs | |
| CN111222332B (en) | Commodity recommendation method combining attention network and user emotion | |
| CN108763362B (en) | Local model weighted fusion Top-N movie recommendation method based on random anchor point pair selection | |
| CN110956497A (en) | Method for predicting repeated purchasing behavior of user of electronic commerce platform | |
| CN113468227B (en) | Information recommendation method, system, equipment and storage medium based on graph neural network | |
| CN111259140B (en) | False comment detection method based on LSTM multi-entity feature fusion | |
| CN108921602B (en) | User purchasing behavior prediction method based on integrated neural network | |
| CN110287329B (en) | E-commerce category attribute mining method based on commodity text classification | |
| CN109033294B (en) | Mixed recommendation method for integrating content information | |
| CN107944911A (en) | A kind of recommendation method of the commending system based on text analyzing | |
| CN112487199A (en) | User characteristic prediction method based on user purchasing behavior | |
| CN111429161A (en) | Feature extraction method, feature extraction device, storage medium, and electronic apparatus | |
| CN114942998B (en) | Knowledge graph neighborhood structure sparse entity alignment method integrating multi-source data | |
| CN111191099A (en) | User activity type identification method based on social media | |
| CN111178902B (en) | Network payment fraud detection method based on automatic feature engineering | |
| CN111985680A (en) | Criminal multi-criminal name prediction method based on capsule network and time sequence | |
| Yu et al. | Graph learning for fake review detection | |
| CN114048754A (en) | Chinese short text classification method integrating context information graph convolution | |
| CN110134764A (en) | A kind of automatic classification method and system of text data | |
| CN107609921A (en) | A kind of data processing method and server | |
| CN114238758B (en) | User portrait prediction method based on multi-source cross-border data fusion | |
| CN111666410B (en) | Emotion classification method and system for commodity user comment text | |
| CN110472680B (en) | Object classification method, device and computer-readable storage medium | |
| CN114461928A (en) | Session recommendation method based on graph neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |