CN103886169A - Link prediction algorithm based on AdaBoost - Google Patents
Link prediction algorithm based on AdaBoost Download PDFInfo
- Publication number
- CN103886169A CN103886169A CN201210553291.8A CN201210553291A CN103886169A CN 103886169 A CN103886169 A CN 103886169A CN 201210553291 A CN201210553291 A CN 201210553291A CN 103886169 A CN103886169 A CN 103886169A
- Authority
- CN
- China
- Prior art keywords
- prediction
- sorter
- sample
- link
- algorithm
- 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.)
- Pending
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a link prediction algorithm based on AdaBoost. The link prediction algorithm is applicable to predicting future communication probability of the communication entity in a current topology structure. When a user inputs the communication relation of a current network, the prediction about whether the next-moment communication entity transmit communication or not can be obtained through a series of calculation. Compared with various common prediction algorithms, the link prediction algorithm has the advantages that the concept of enhancing weak learning methods into strong learning methods in Boosting methods is applied to link prediction; the link prediction algorithm is high in flexibility and low in false alarm rate, algorithm recalling rate can be increased evidently, and accuracy of calculation results can be kept.
Description
Technical field
The present invention relates to Internet technology, be specifically related to a kind of implementation method of link prediction.
Background technology
Link prediction is using link as the application of excavating object during link excavates.Main prediction has existed but the following possibility that produces link between the node of not yet found link and not yet link.Along with some link prediction algorithms start to be applied at commercial field, associated research has become a popular domain, and wherein the link prediction algorithm research based on topological diagram is operated in and has been subject in recent years extensive attention.For example Facebook adopts the friends of the method predictive user based on RWR (Random Walk with Restart), improves accordingly the success ratio of friend recommendation.
The link prediction algorithm of topological diagram Network Based mainly comprises the similarity based on node neighbours, based on maximal possibility estimation and based on three types such as probability models.Representative algorithm comprises common neighbours (CommonNeighbors) algorithm, the Katz algorithm based on path similarity and the RWR algorithm based on random walk similarity based on local message similarity.Wherein, the link prediction algorithm research based on node neighbours similarity early, has been obtained widespread use in real work.It is another kind of that to obtain the actual method of applying be the link prediction algorithm based on random walk.The basic thought of this class algorithm is all to all possible line ordering that is combined into of node in figure, selects most probable wherein to appear at node in new figure to (i.e. limit in figure).But nearly one or two years no matter be in the improvement to existing algorithm, or aspect proposition new algorithm, all do not have breakthrough achievement, the recall rate of the link prediction algorithm based on topological is still lower.
Summary of the invention
The object of this invention is to provide a kind of link prediction algorithm based on AdaBoost.Use embodiment provided by the invention, can be to the node that in current network topological diagram, link may occur in the future to predicting.
In order to overcome the lower problem of link prediction algorithm ubiquity recall rate of topological structure Network Based of current main-stream.By we research find, existing main flow link prediction method predict the outcome and not exclusively crossing, utilize arithmetic result stack improve recall rate.But directly cumulative summation infeasible, because can reduce total arithmetic accuracy.Consider accordingly to adopt Boosting method to make improvements.First regard link prediction problem as two classification problems, to each limit that may exist in next moment network (node to), its classification results is two classes: exist or do not exist.Next use Boosting method and promote weak learning algorithm and obtain the thought of strong learning algorithm by error feedback, select some link prediction algorithms as Weak Classifier according to certain principle, propose and realized a new link prediction method based on AdaBoost algorithm.
The step of the method comprises:
Read prediction training sample and prediction test sample book;
For prediction training sample is enclosed the label value of its true place class;
For the weight of each sample is composed initial value;
Choose some link prediction algorithms as Weak Classifier;
Use each sorter to classify for training sample does;
Calculate the ballot weight of each sorter;
Use each sorter to classify for the sample in prediction test set does;
Be the sample ballot in prediction test set by the classification results of above-mentioned each sorter, make final prediction;
Output predicts the outcome to sample in prediction test set;
Finally, implement the present invention and there is following beneficial effect:
The beneficial effect of the embodiment of the present invention is, Boosting thought is applied among link prediction, for existing various algorithms most in use, there is higher sensitivity and lower rate of false alarm, can, in significantly improving algorithm recall rate, keep the correctness of result of calculation.
Brief description of the drawings
Accompanying drawing is the algorithm flow that the present invention improves a kind of link prediction algorithm based on AdaBoost of existing link prediction algorithm proposition.
Embodiment
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described, so that those skilled in the art understands the present invention better.
In the present embodiment, as shown in the figure, provide the algorithm flow of an optimization:
Step 101, read prediction training sample and prediction test sample book;
For prediction training sample and training test sample book, read its information generating network topological diagram.
Step 102, for prediction training sample enclose the label value of its true place class;
The prediction that is m for one group of length training set C.Ω represents x
ithe set of the types value being classified.For x
iif it appears in the figure of next time period really, y
i=1, otherwise, y
i=-1.
Step 103, be each sample weight compose initial value;
The weight initial value of each sample equates, is the inverse of whole sample length, is 1/m.
Step 104, choose some link prediction algorithms as Weak Classifier;
Choose the similarity based on node neighbours according to the complementary principle that predicts the outcome, based on maximal possibility estimation and the method based on three types link prediction such as probability models as Weak Classifier.
Step 105, use each sorter to classify for training sample does;
To each prediction algorithm t, use algorithm for a value of every a pair of node calculating in C, then according to this value to node to carrying out descending sort, choose a front y node pair, form set P, represent that algorithm t thinks that these nodes exist in the figure in next moment meeting, remaining formation set Q, represents that algorithm t thinks that these nodes to not existing in the figure in next moment.Y is the right number of node of gathering in the figure that actually exists in next time period in C.Regard prediction algorithm t as a Weak Classifier t, what t made is assumed to be h
t.If x
i∈ P, h
t(x
i)=1, otherwise, if x
i∈ Q, h
t(x
i)=-1.
Step 106, calculate the ballot weight of each sorter;
Carry out T circulation, t=1 .., T: first circulation time is each time the current error rate of each classifier calculated.For each sample, by sorter t to its classification compared with type under itself, if inconsistent, in the error rate of this sorter, add the weight of this sample.Calculate and the sorter of the rate minimum that locates errors as current sorter.If but error rate is greater than 1/2, just stop algorithm.Error rate is normalized, as the ballot weight of current sorter t.Upgrade the weight of each sample, if this sample by current sorter mis-classification, its weight rise.X comparatively speaking
iif correctly classified so, its weight has just reduced.After T circulation, obtain the ballot weight of each sorter.
Step 107, use each sorter to classify for the sample in prediction test set does;
In prediction test set D, use e
jrepresent the each node pair in D, n is the right number of all nodes in D.For every kind of prediction algorithm t, be that in prediction test set D, every a pair of node calculates a value, then carry out according to this value node, to carrying out descending sort, choosing the individual node pair of front m ', form set P ', remaining formation set Q '.M ' is the right number of node of gathering in the figure that actually exists in next time in D.If e
j∈ P ', h
t(e
j)=1, otherwise, if e
j∈ Q ', h
t(e
j)=-1.
Step 108, make final prediction for test sample book;
For each e
j, by each Weak Classifier t, it is voted.If sorter t thinks e
jin next time chart, exist, e
jweight add the ballot weight of this sorter.If sorter t thinks e
jin next time chart, not existing, is e
jweight deduct the ballot weight of this sorter.At all sorters to e
jafter having voted, if e
jweight for just, predict e
jcan in the figure in next moment, exist.Otherwise, e
jweight for negative, predict e
jcan in the figure in next moment, not exist.
Step 109, output predict the outcome to prediction test sample book
Export predicting the outcome
Although above the illustrative embodiment of the present invention is described; so that the technician of this technology neck understands the present invention; but should be clear; the invention is not restricted to the scope of embodiment; to those skilled in the art; as long as various variations appended claim limit and definite the spirit and scope of the present invention in, these variations are apparent, all utilize innovation and creation that the present invention conceives all at the row of protection.
Claims (1)
1. the link prediction algorithm based on AdaBoost: it is characterized in that, first read prediction training sample and prediction test sample book; For prediction training sample is enclosed the label value of its true place class; For the weight of each sample is composed initial value; Choose some link prediction algorithms as Weak Classifier according to arithmetic result principle of complementarity; Use each sorter to classify for training sample does; Cycle calculations obtains all sorter ballot weights, to each circulation, whether correctly calculates the current error rate of each sorter according to classification results, selects the sorter of error rate minimum, calculates its ballot weight, and all samples are carried out to weight upgrading.After finishing, circulation obtains the ballot weight of each sorter; Use each sorter to classify for the sample in prediction test set does; Be the sample ballot in prediction test set by the classification results of above-mentioned each sorter, final vote result is that positive sample is prediction it can link in the future, and voting results are that negative sample is and predicts that it can not link in the future again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210553291.8A CN103886169A (en) | 2012-12-19 | 2012-12-19 | Link prediction algorithm based on AdaBoost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210553291.8A CN103886169A (en) | 2012-12-19 | 2012-12-19 | Link prediction algorithm based on AdaBoost |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103886169A true CN103886169A (en) | 2014-06-25 |
Family
ID=50955060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210553291.8A Pending CN103886169A (en) | 2012-12-19 | 2012-12-19 | Link prediction algorithm based on AdaBoost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103886169A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104363092A (en) * | 2014-09-25 | 2015-02-18 | 电子科技大学 | Fixed-range device authentication based on audio physical fingerprints |
| CN106959967A (en) * | 2016-01-12 | 2017-07-18 | 中国科学院声学研究所 | A kind of training of link prediction model and link prediction method |
| CN108154071A (en) * | 2016-12-05 | 2018-06-12 | 北京君正集成电路股份有限公司 | Detector training method and device, the detection method and device of pedestrian's moving direction |
| US10572501B2 (en) | 2015-12-28 | 2020-02-25 | International Business Machines Corporation | Steering graph mining algorithms applied to complex networks |
-
2012
- 2012-12-19 CN CN201210553291.8A patent/CN103886169A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104363092A (en) * | 2014-09-25 | 2015-02-18 | 电子科技大学 | Fixed-range device authentication based on audio physical fingerprints |
| CN104363092B (en) * | 2014-09-25 | 2018-06-19 | 电子科技大学 | The device authentication based on audio physical fingerprint under the conditions of spacing |
| US10572501B2 (en) | 2015-12-28 | 2020-02-25 | International Business Machines Corporation | Steering graph mining algorithms applied to complex networks |
| CN106959967A (en) * | 2016-01-12 | 2017-07-18 | 中国科学院声学研究所 | A kind of training of link prediction model and link prediction method |
| CN108154071A (en) * | 2016-12-05 | 2018-06-12 | 北京君正集成电路股份有限公司 | Detector training method and device, the detection method and device of pedestrian's moving direction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11659050B2 (en) | Discovering signature of electronic social networks | |
| Guo et al. | Deep information fusion-driven POI scheduling for mobile social networks | |
| CN104134159B (en) | A kind of method that spread scope is maximized based on stochastic model information of forecasting | |
| TWI726341B (en) | Sample attribute evaluation model training method, device, server and storage medium | |
| CN102567391B (en) | Method and device for building classification forecasting mixed model | |
| CN112631717B (en) | Asynchronous reinforcement learning-based network service function chain dynamic deployment system and method | |
| US20210383205A1 (en) | Taxonomy Construction via Graph-Based Cross-domain Knowledge Transfer | |
| CN112651534B (en) | Method, device and storage medium for predicting resource supply chain demand | |
| CN107292390A (en) | A kind of Information Propagation Model and its transmission method based on chaology | |
| CN107886160B (en) | BP neural network interval water demand prediction method | |
| CN116842459B (en) | Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning | |
| CN115270007B (en) | POI recommendation method and system based on mixed graph neural network | |
| CN105471647A (en) | Power communication network fault positioning method | |
| CN103886169A (en) | Link prediction algorithm based on AdaBoost | |
| CN107453921A (en) | Smart city system artificial intelligence evaluation method based on nonlinear neural network | |
| CN115049397A (en) | Method and device for identifying risk account in social network | |
| CN115456093A (en) | High-performance graph clustering method based on attention-graph neural network | |
| CN113868537B (en) | Recommendation method based on multi-behavior session graph fusion | |
| CN106257507A (en) | The methods of risk assessment of user behavior and device | |
| CN104092503A (en) | Artificial neural network spectrum sensing method based on wolf pack optimization | |
| CN107644268B (en) | Open source software project incubation state prediction method based on multiple features | |
| CN106503794A (en) | A kind of gear case of blower method for predicting residual useful life | |
| CN116384240A (en) | Server energy consumption prediction method, device and storage medium | |
| CN114842247B (en) | Characteristic accumulation-based graph convolution network semi-supervised node classification method | |
| CN116307078A (en) | Account label prediction method and device, storage medium and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140625 |