(19)
(11)EP 3 644 232 A1

(12)EUROPEAN PATENT APPLICATION
published in accordance with Art. 153(4) EPC

(43)Date of publication:
29.04.2020 Bulletin 2020/18

(21)Application number: 18889062.8

(22)Date of filing:  16.08.2018
(51)Int. Cl.: 
G06K 9/62  (2006.01)
(86)International application number:
PCT/CN2018/100758
(87)International publication number:
WO 2019/114305 (20.06.2019 Gazette  2019/25)
(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
Designated Extension States:
BA ME
Designated Validation States:
KH MA MD TN

(30)Priority: 12.12.2017 CN 201711322274

(71)Applicant: ALIBABA GROUP HOLDING LIMITED
George Town, Grand Cayman (KY)

(72)Inventors:
  • ZHOU, Shuheng
    Hangzhou, Zhejiang 31121 (CN)
  • ZHU, Huijia
    Hangzhou, Zhejiang 31121 (CN)
  • ZHAO, Zhiyuan
    Hangzhou, Zhejiang 31121 (CN)

(74)Representative: Goddar, Heinz J. 
Boehmert & Boehmert Anwaltspartnerschaft mbB Pettenkoferstrasse 22
80336 München
80336 München (DE)

  


(54)METHOD AND DEVICE FOR CLASSIFYING SAMPLES TO BE ASSESSED


(57) Implementations of this specification provide a method for classifying samples to be assessed. The method includes: obtaining samples T to be assessed and sample feature Ft; select a certain quantity N of example samples from a classification sample database; obtaining feature similarity SIMi of the samples T to be assessed and each example sample i among the N example samples; obtaining sample quality Qi of each example sample i; determining comprehensive similarity Si between samples T to be assessed and each example sample i based on at least a difference value ri between feature similarity SIMi and sample quality Qi; and determining whether samples T to be assessed fall in a category in the classification sample database based on comprehensive similarity Si. Corresponding devices are also provided. With the described method and devices, the samples to be assessed can be classified more effectively and accurately.




Description

TECHNICAL FIELD



[0001] One or more implementations of the present specification relate to the field of computer technologies, and in particular, to sample classification and identification.

BACKGROUND



[0002] As the Internet upgrades, a wide variety of information and content are generated on the network every day. In many cases, these information and content need to be identified and classified. For example, many network platforms generate a large amount of junk information, advertising information, etc. To ensure user experience, junk information and advertising information need to be identified and filtered. For another example, to improve the network environment, it is also necessary to identify and classify content of the network that contains pornography, violence or that violates laws and regulations.

[0003] To identify and classify the network content, the method of establishing a classification sample database is usually used. For example, an advertisement "black sample" database can be created for advertising information and store collected example samples, which are also referred to as black samples. The network content to be assessed is compared with the black samples in the black sample database to determine whether the network content to be assessed fall in the same category in the classification sample database according to the similarity of the comparison, that is, whether the network content is also advertisement content.

[0004] Typically, the sample database contains a large quantity of example samples. These samples are usually collected manually and therefore vary in quality. Some example samples are of low quality, and have poor generalization ability. Therefore, the content to be assessed does not fall in the same category as the example sample even though the content has a high similarity with the example sample. This brings much difficulty in classifying and assessing samples.

[0005] Therefore, improved solution is needed to assess and classify the content to be assessed and samples more effectively.

SUMMARY



[0006] One or more implementations of this specification describe a method and a device. Similarity between a sample to be assessed and an example sample is assessed more effectively and more accurately by introducing consideration of sample quality of the example sample during assessment.

[0007] According to a first aspect, a method for classifying samples to be assessed is provided, including: obtaining samples T to be assessed and sample feature Ft of sample T to be assessed; selecting a first quantity N of example samples from a classification sample database; obtaining feature similarity SIMi of the samples T to be assessed and each example sample i among the N example samples, where feature similarity SIMi is determined based on sample feature Ft of samples T to be assessed and sample feature Fi of each example sample i; obtaining sample quality Qi of each example sample i; determining comprehensive similarity Si between sample T to be assessed and each example sample i based on at least a difference value ri between feature similarity SIMi and sample quality Qi; and determining, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.

[0008] In an implementation, the selecting the first quantity N example samples from a classification sample database includes: calculating feature similarity between samples T to be assessed and each second quantity M example samples based on sample feature Ft of sample T to be assessed and sample features of the second quantity M example samples in the classification sample database, where the second quantity M is greater than the first quantity N; and selecting the first quantity N example samples from the second quantity M example samples based on feature similarity between the sample to be assessed and each second quantity M example samples.

[0009] In an implementation, the selecting the first quantity N of example samples from a classification sample database includes selecting the first quantity N of example samples from the classification sample database based on the sorting result of sample quality of each sample in the classification sample database.

[0010] According to one implementation, feature similarity SIMi is determined by normalizing the distance between sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.

[0011] In an implementation, determining comprehensive similarity Si between sample T to be assessed and each example sample i includes determining comprehensive similarity Si as Si = a + b ri c, where a + b = 1, and c is a coefficient associated with sample quality Qi.

[0012] In an implementation, in the case of ri > = 0, c = 1/(1-Qi) and in the case of ri < 0, c = 1/Qi.

[0013] According to one implementation, the method above further includes :determining a total similarity score of the sample to be assessed based on comprehensive similarity Si between the sample to be assessed and each example sample i.

[0014] In an implementation, the determining a total similarity score of the sample to be assessed includes: if at least one ri > = 0, determining the total similarity score as the maximum value among comprehensive similarities Si between sample T to be assessed and each example sample i; or otherwise, determining the total similarity score as the minimum value among comprehensive similarities Si between sample T to be assessed and each example sample i.

[0015] In an implementation, the determining a total similarity score of sample to be assessed includes determining the total similarity score as the average value of comprehensive similarities Si between sample T to be assessed and each example sample i.

[0016] According to a second aspect, a device for classifying samples to be assessed is provided, including: a sample acquisition unit, configured to obtain sample T to be assessed and sample feature Ft of sample T to be assessed; a selection unit, configured to select the first quantity N of example samples from a classification sample database; a first acquisition unit, configured to obtain feature similarity SIMi between sample T to be assessed and each example sample i of the N example samples, where feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i; a second acquisition unit, configured to obtain sample quality Qi of each example sample i; a processing unit, configured to determine a comprehensive similarity Si between sample T to be assessed and each example sample i based on at least a difference value ri between feature similarity SIMi and sample quality Qi; and a classification unit, configured to determine, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.

[0017] According to a third aspect, a computer readable storage medium is provided, where the medium stores a computer program, and when the computer program is executed in a computer, the computer is enabled to perform the method of the first aspect.

[0018] According to a fourth aspect, a computing device is provided, including a memory and a processor, where the memory stores executable code, and when the processor executes the executable code, the method of the first aspect is implemented.

[0019] When the method and the device provided in the implementations of this specification are used, the feature similarity between the sample to be assessed and the example sample and the sample quality of the example sample are comprehensively considered to determine the comprehensive similarity between the sample to be assessed and the example sample. Based on the comprehensive similarity, the sample to be assessed is classified, thereby reducing or avoiding the adverse impact of the varied sample quality on the assessment results and making it more effective and more accurate to determine the category of the sample to be assessed.

BRIEF DESCRIPTION OF DRAWINGS



[0020] To describe the technical solutions in the implementations of the present disclosure more clearly, the following briefly introduces the accompanying drawings required for describing the implementations. Apparently, the accompanying drawings in the following description are merely some implementations of the present disclosure, and a person of ordinary skill in the field may still derive other drawings from these accompanying drawings without creative efforts.

FIG. 1 is a schematic diagram illustrating an application scenario of an implementation disclosed in this specification;

FIG. 2 is a flowchart illustrating a method, according to one implementation;

FIG. 3 is a flowchart illustrating selection of a certain quantity of example samples, according to one implementation;

FIG. 4 is a flowchart illustrating selection of a certain quantity of example samples, according to another implementation;

FIG. 5 is a flowchart illustrating selection of a quantity of example samples, according to another implementation; and

FIG. 6 is a schematic block diagram illustrating a classification device, according to one implementation.


DESCRIPTION OF IMPLEMENTATIONS



[0021] The solution provided in this specification is described below with reference to the accompanying drawings.

[0022] FIG. 1 is a schematic diagram illustrating an application scenario of an implementation disclosed in this specification. In FIG. 1, a processing platform obtains a sample to be assessed and sample information of example samples from a sample database. The sample information includes sample features of example samples and sample quality of example samples. The processing platform then determines comprehensive similarity between the sample to be assessed and the example samples based on feature similarity between the sample to be assessed and each example samples and sample quality of the example samples. The described processing platform can be any platforms with computing and processing capabilities, such as a server. The described sample database can be sample database created by collecting samples and is used to classify or identify samples, including a plurality of example samples. Although the sample database shown in FIG. 1 is stored in an independent database, it can be understood that the sample database can also be stored in the processing platform. By using assessment methods in the implementations, the processing platform uses the sample quality of the example samples as a factor in determining the comprehensive similarity between the sample to be assessed and the example samples. Therefore, reverse impact of the varied sample quality of the example samples on the assessment results is reduced or avoided.

[0023] The following describes in detail the method the processing platform used to classify samples to be assessed. FIG. 2 is a flowchart illustrating a method, according to one implementation. The process can be performed by a processing platform with computing capability, such as a server as shown in FIG. 1. As shown in FIG. 2, the method includes the following steps:

Step S21: Obtain sample T to be assessed and sample feature Ft of sample T to be assessed.

Step S22: Select the first quantity N of example samples from classification sample database.

Step S23: Obtain feature similarity SIMi between sample T to be assessed and each example sample i of the first quantity N of example samples, where feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.

Step S24: Obtain sample quality Qi of each example sample i. Sample quality Qi corresponds to a similarity threshold, that is, a historical assessment sample whose feature similarity with the example sample i exceeds the similarity threshold is determined as a specific category in a certain ratio.

Step S25: Determine comprehensive similarity Si between sample T to be assessed and each example sample i based on at least a difference value ri between feature similarity SIMi and sample quality Qi.

Step S26: Determine, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.



[0024] First, in step S21, sample T to be assessed and sample feature Ft of the sample to be assessed are obtained. It can be understood that sample T to be assessed can be various objects to be assessed and classified, such as a text, a picture, and code. In an implementation, the processing platform needs to automatically detect, assess, or classify various content uploaded onto a network. In this case, obtaining sample T to be assessed includes capturing the sample to be assessed from the network. For example, when the processing platform needs to filter advertisement images on the network. samples of images to be assessed can be captured from the network. In another implementation, obtaining sample T to be assessed includes receiving sample T to be assessed, that is, the processing platform analyzes and assesses the received samples to be assessed. For example, after a mobile phone communication system receives a message, the mobile phone communications system needs to determine whether the message is a junk message. In this case, the message can be sent to the processing platform for message classification. The processing platform then assesses and classifies the received message.

[0025] For sample T to be assessed, sample feature Ft can be extracted. Sample feature Ft is extracted for machine learning and analysis and is used to identify different samples. In the existing technology, many models can be used to extract features of various samples to implement comparison and analysis. For example, for a picture sample, sample features can include the following: quantity of pixels, mean gray value, median gray value, quantity of sub-regions, sub-region size, sub-region mean gray value, etc. For text samples, sample features can include words in text, quantity of words, word frequency, etc. For other types of samples, there are corresponding feature extraction methods. Generally, sample features include a plurality of feature elements, and therefore, sample features can be represented as a feature vector composing a plurality of feature elements.

where Ti is the feature elements of the sample to be assessed.

[0026] In addition, in step S22, select the first quantity N of example samples from the classification sample database.

[0027] It can be understood that the classification sample database is established by collecting samples in advance and is used to classify, compare and identify samples. The database contains a plurality of example samples. For example, a sample database of advertisement pictures contains a large quantity of example advertisement pictures, and a sample database of junk messages contains a plurality of example junk messages.

[0028] In an implementation, a quantity of example samples contained in the sample database is relatively small for example, the quantity of example samples is less than a certain threshold (for example, 100). In this case, all example samples in the sample database may be used for performing subsequent steps S23-S25. That is, the first quantity N in step S22 is the quantity of all example samples in the classification sample database.

[0029] In another implementation, the quantity of example samples contained in the classification sample database is large. For example, the quantity of example samples is greater than a certain threshold (for example, 200). Alternatively, content of the example samples in the sample database is not concentrated. For example, although all samples stored in the sample database of advertisement pictures are advertisement pictures, content of these pictures differs because these pictures may contain either people or items or scenery. In this case, the example samples in the sample database can be filtered to determine a quantity N of more targeted example samples for further processing.

[0030] Many ways can be used to determine a certain quantity N example samples from the classification sample database. FIG. 3 is a flowchart illustrating selection of a quantity of example samples based on one implementation. As shown in FIG. 3, first in step S31, sample feature Fi of each example sample i in a classification sample database are obtained. It can be understood that, in correspondence with the sample to be assessed, sample feature Fi of each example sample i may similarly be represented by a feature vector.



[0031] In step S32, feature similarity SIMi between sample T to be assessed and each example sample i is calculated based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.

[0032] In an implementation, the distance di between sample T to be assessed and the example sample i is first calculated, and the distance di is normalized to obtain feature similarity SIMi. It can be understood that because both sample T to be assessed and the example sample i can be represented in the form of a feature vector, various algorithms can be used to calculate the distance between the two vectors as the distance di. For example, the Euclidean distance between feature vector Ft of sample T to be assessed and feature vector Fi of the example sample i may be calculated as the distance di by using conventional mathematical methods. Alternatively, the Mahalanobis distance or the Hamming distance, etc. between Ft and Fi may be calculated as the distance di between sample T to be assessed and the example sample i. Then, the distance can be normalized to obtain feature similarity SIMi. In one example, the distance is normalized by using the following equation:



[0033] To make the value of SIMi is between 0 and 1. It can be understood that other normalization methods can also be used.

[0034] In an implementation, feature similarity SIMi between sample T to be assessed and the example sample i is determined based on cosine similarity between feature vector Ft and feature vector Fi. In this method, the cosine value of the angle between feature vector Ft and feature vector Fi is used to directly determine feature similarity SIMi between 0 and 1. A person skilled in the field may also use other algorithms to determine the feature similarity based on the respective feature vectors of sample T to be assessed and the sample feature i.

[0035] Therefore, in step S32, feature similarity SIMi between sample T to be assessed and each example sample i in the sample database is calculated. Next, in step S33, a certain quantity N of example samples are selected from the classification sample database based on each of calculated feature similarity SIMi.

[0036] In an implementation, feature similarities SIMi between sample T to be assessed and each example sample i are first sorted, and the N example samples are selected based on the sorting result.

[0037] In one example, the N example samples with the highest feature similarity to sample T to be assessed are selected. For example, N can be 10 or 20. Of course, example samples whose feature similarities are sorted in a predetermined range, such as between the 5th and the 15th, are selected. The method for selecting example samples can be set as needed.

[0038] In another example, exceptional values of the feature similarities that deviate from the predetermined range are first removed, and the N example samples with the highest feature similarities to samples T to be assessed are selected from the sorting result after the exceptional values are removed.

[0039] In another implementation, the certain quantity N is not predetermined. Correspondingly, an example sample with feature similarity in a predetermined range can be selected as a selected example sample. For example, a threshold can be predetermined to select example samples with feature similarity SIMi that are greater than the threshold.

[0040] As such, a certain quantity (N) of example samples are selected from the classification sample database, and the selected example samples have a higher feature similarity to the sample to be assessed, that is, features of the selected example samples are more similar to features of the sample to be assessed. Therefore, the selected example samples are more targeted and more advantageous for the accuracy of subsequent processing results.

[0041] The process of selecting example samples can also be implemented in other ways. FIG. 4 is a flowchart diagram illustrating selection of a certain quantity (the first quantity N) of example samples, according to another implementation. As shown in Figure 4, first in step S41, M (the second quantity) example samples are selected from a classification sample database to obtain sample feature Fi of each example sample i of the M example samples. It can be understood that the second quantity M of example samples are initially selected example samples, and the quantity M is greater than the previous first quantity N. In an implementation, the next step is performed by randomly selecting M example samples from the classification sample database. Alternatively, the most recently used M example samples are selected from the classification sample database to perform the next step. The second quantity M can also be determined based on a predetermined ratio, for example, 50% of the total quantity of all example samples in the classification sample database.

[0042] Next, in step S42, feature similarity SIMi between sample T to be assessed and each example sample i is calculated based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i of the selected M example samples. For the method for calculating feature similarity SIMi in the present step, references can be made to the description of step S32 in FIG. 3. Details are omitted here for simplicity.

[0043] Then in step S43, the first quantity N of example samples is further selected from the M example samples based on calculated feature similarities SIMi. For the method for selecting the N example samples from more example samples based on feature similarity SIMi in the present step, references can be made to descriptions of step S33 in FIG. 3. Details are omitted here for simplicity.

[0044] As can be seen from the comparison between the implementation in FIG. 4 and the implementation in FIG. 3, the implementation in FIG. 4 differs from the implementation in FIG. 3 in that the M example samples are initially selected from the classification sample database to calculate the feature similarity between the sample to be assessed and the M example samples, and then the N example samples are further selected from the M example samples based on the feature similarity. This is particularly applicable when the quantity of example samples in the classification sample database is very large. In this case, the calculation cost of calculating the feature similarity between each example sample in the classification sample database and the sample to be assessed (step S32) is still high and the implementation in FIG. 4 can be adopted.

[0045] In practice, the N example samples that are finally selected are typically in multiples of ten, such as 10, 20, and 50. Therefore, the implementation in FIG. 3 can be adopted if the quantity of the example samples in the classification sample database is in the level of thousand. If the quantity of example samples in the classification sample database is very large, for example, there are tens of thousands or even hundreds of thousands of example samples, to speed up processing, the method in the implementation in FIG. 4 can be adopted. First, a portion of M example samples are selected from the classification sample database. For example, the quantity of the M example samples may be thousands or hundreds. Then tens of example samples are further selected based on the feature similarity for subsequent processing.

[0046] FIG. 5 is a flowchart illustrating selection of a quantity of example samples, according to another implementation. As shown in FIG. 5, in step S51, sample quality Qi of each example sample i in a classification sample database is obtained.

[0047] Sample quality Qi is used to measure the generalization ability of an example sample. The example sample corresponds to a similarity threshold that a historical assessment sample whose feature similarity to the example sample i exceeds the similarity threshold is determined in a ratio to fall in the same category as the classification sample database. In one example, a historical assessment sample whose feature similarity to the example sample i exceeds the similarity threshold is considered as falling in the same category as the classification sample database. Therefore, when the feature similarity between the sample to be assessed and the example sample exceeds Qi, there is greater reason to believe that the sample to be assessed and the example sample fall in the same category. For example, for an example sample in a junk message sample database, if the sample quality is 0.6, this means if the feature similarity score of a sample to be assessed exceeds 0.6, there is a great probability that the sample to be assessed is also a junk message. For another example, for an example sample in an advertisement picture sample database, if the sample quality is 0.8, this means if the feature similarity score of a sample to be assessed exceeds 0.8, there is a great probability that the sample to be assessed is also an advertisement picture. Generally, the lower the value of sample quality Q, the stronger the generalization ability of the sample .

[0048] Sample quality Qi can be determined in several ways. In an implementation, the sample quality of each example sample is determined manually and example samples are stored in the classification sample database. In another implementation, sample quality Qi is determined based on historical data of sample assessment classification. Specifically, the sample quality of a certain example sample is determined by obtaining feature similarities between a plurality of existing samples and the example sample , and the final assessment results of the plurality of existing samples. More specifically, the lowest feature similarity score among respective feature similarity scores between the example sample and the existing samples that are finally identified as falling in the same category can be determined as the sample quality score of the example sample. For example, for example sample k, five historical assessment samples were compared with k in historical records. Assume that the result of the comparison shows that the feature similarity scores of these five historical assessment samples to sample k are SIM1 = 0.8, SIM2 = 0.6, SIM3 = 0.4, SIM4 = 0.65, SIM5 = 0.7 respectively. Finally, the historical assessment samples whose feature similarity values are 0.6 and 0.4 are not considered to be fall in the same category as sample k, and other historical assessment samples are considered to be in the same category. In this case, sample quality Q of sample k can be considered to be 0.65, that is, the lowest value among the feature similarities between sample k and the three historical assessment samples that fall in the same category.

[0049] In an implementation, in step S51, sample quality Qi of each example sample i in a classification sample database is calculated by the historical records. In another implementation, the sample quality has been pre-calculated and is stored in the sample database. In step S51, sample quality Qi of each example sample i is read.

[0050] Next, in step S52, a certain quantity N of example samples are selected from the classification sample database based on the sorting result of the sample quality Qi of each example sample i described above. In an implementation, N example samples with the lowest values of Qi are selected from the classification sample database. In another implementation, a value of N is not specified in advance. In this case, example samples whose values of sample quality Qi are below a certain threshold can be selected. In this way, N example samples with a strong generalization ability are selected from the classification sample database for further processing.

[0051] In addition to the methods shown in FIG. 3, FIG. 4, and FIG. 5, a person skilled in the field can use a similar method to select the first quantity N of example samples from the classification sample database after reading this specification. therefore, step S22 in FIG. 2 is performed.

[0052] Referring back to FIG. 2, on the basis of selecting the N example samples, in step S23, feature similarity SIMi between sample T to be assessed and each example sample i among the N example samples are obtained, where feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.

[0053] It can be understood that if the N example samples are selected in the methods shown in FIG. 3 or FIG. 4, feature similarities SIMi between sample T to be assessed and all example samples/the M example samples have been calculated during the selection process. Correspondingly, in step S23, only the feature similarities between sample T to be assessed and the N selected example samples needs to be read from the calculation result.

[0054] If other methods are used to select the N example samples, then in step S23, feature similarity SIMi between sample T to be assessed and each example sample i is calculated based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i in the N selected example samples. For the calculation method, references can be made to the description of step S32 in FIG. 3. Details are omitted here for simplicity.

[0055] In addition, in step S24, sample quality Qi of each of the N example samples selected is obtained.

[0056] It can be understood that if the N example samples are selected in the method shown in FIG. 5, the sample quality of all example samples has been obtained during the selection process. Correspondingly, in step S24, only the sample quality of the N selected example samples needs to be read from all results.

[0057] If the N example samples are selected in other methods, in step S24, obtain sample quality of the N example samples. For the method for obtaining the sample quality, references can be made to the description of step S51 in FIG. 5. Details are omitted here for simplicity.

[0058] On the basis of obtaining feature similarity SIMi between each example sample i and the sample to be assessed, and sample quality Qi of each example sample, in step S25, comprehensive similarities Si between sample to be assessed and each example sample i are obtained based on at least a difference value ri between feature similarity SIMi and sample quality Qi.

[0059] In an implementation, comprehensive similarity Si is determined to be Si = a + b ri c, where a and b are constants, a + b = 1, and c is a coefficient associated with sample quality Qi.

[0060] For example, in one example, Si = 0.8 + 0.2 ri/2Qi;
In another example, Si = 0.7 + 0.3 ri/Qi.

[0061] In an implementation, parameter c is set to be different values for different values of ri. For example, in the case of ri > = 0, c = 1/(1-Qi) and in the case of ri < 0, c = 1/Qi.

[0062] In an example, the calculation form of Si is as follows:



[0063] In the previous equation, in the case of ri > = 0, make c = 1/(1-Qi). Therefore, ri/(1-Qi) is not greater than 1, so Si is not greater than 1. In addition, the effect of the difference value ri between the feature similarity SIMi and the sample quality Qi can be better measured. If a value of Qi is relatively large or even closer to 1, a margin (1-Qi) of the difference value ri will be very small. In this case, Si should be calculated by considering the ratio of difference value ri to its possible margin. In the case of ri < 0, c can be directly set to be 1/Qi , and Si can be calculated by considering the ratio of difference value ri to Qi.

[0064] In the process of calculating the comprehensive similarity, because the sample quality, the difference value between the feature similarity and the sample quality are comprehensively considered, the calculated comprehensive similarity can more objectively reflect the probability that the sample to be assessed and the example sample fall in the same category, and is less affected by the sample quality of the example sample. For example, assume there are two example samples A and B, and sample quality is QA = 0.4 and QB = 0.8 respectively. Assume that feature similarity between sample T to be assessed and sample A and feature similarity between sample T to be assessed and sample B are both 0.7. In such a situation, if only consider the feature similarity, it is generally considered that the sample to be assessed is either similar or dissimilar to both the example samples because the feature similarity between sample T to be assessed and sample A and the feature similarity between sample T to be assessed and sample B are the same. If the method in the previous implementation is used, for example, algorithm of equation 1 is used, comprehensive similarity SA = 0.95 between the sample to be assessed and sample A, and comprehensive similarity SB = 0.8875 between the sample to be assessed and sample B are obtained. The comprehensive similarity shows that the similarity between the sample to be assessed and sample A is different from the similarity between the sample to be assessed and sample B. the sample quality score of example sample A is only 0.4, and feature similarity between the sample to be assessed and sample A is much greater than the threshold requirement for falling in the same category, so comprehensive similarity with sample A is significantly higher. Therefore, the comprehensive similarity obtained this way can more objectively reflect the probability that the sample to be assessed and the example sample fall in the same category.

[0065] As such, in step S25, comprehensive similarities between sample T to be assessed and the N example samples are respectively calculated. Further, in step S26, it can be determined whether sample T to be assessed falls in the category of the classification sample database based on the comprehensive similarity Si.

[0066] In an implementation, obtained N comprehensive similarities Si are sorted to determine the highest comprehensive similarity score. The highest comprehensive similarity score is compared with a predetermined threshold, and if the value is greater than the threshold, sample T to be assessed is considered to fall in the same category as the classification sample database.

[0067] In an implementation, total similarity score of the sample to be assessed is determined based on the N comprehensive similarities between sample T to be assessed and the N example samples, and whether sample T to be assessed falls in the category of the classification sample database is determined based on the total similarity score. The total similarity score is used to measure the degree of similarity between the sample to be assessed and the entire example sample set, or the degree of similarity between the sample to be assessed and the entire classification sample database, and the probability of falling in the same category.

[0068] In an implementation, an average value of comprehensive similarity SIMi between sample T to be assessed and each example sample i is calculated, and the average value is determined as the previous total similarity score.

[0069] In another implementation, if at least one of ri among N difference values ri that correspond to the N example samples is greater than or equal to 0, the total similarity score is determined as the maximum value among the comprehensive similarities between sample T to be assessed and the N example samples. Otherwise, the total similarity score is determined as the minimum value among the comprehensive similarities between sample T to be assessed and the N example samples.

[0070] Because sample quality difference of each example sample is considered in determining the total comprehensive score, the sample to be assessed can be determined by setting an appropriate total score threshold in advance. Correspondingly, in step S26, the total similarity score is compared with the predetermined total score threshold, and if the total similarity score of the sample to be assessed is greater than the predetermined total score threshold, the sample to be assessed can be determined as falling in the category of the classification sample database. For example, if the sample to be assessed is a received message, as long as the total similarity score of a junk message sample database is greater than the predetermined threshold, the message is also a junk message.

[0071] According to the method in the previous implementation, the feature similarity between the sample to be assessed and the example sample, and the sample quality of the sample to be assessed are comprehensively considered to determine the comprehensive similarity score of the sample to be assessed and the example sample. Therefore, the adverse impact of varied sample quality on the assessment results is reduced or avoided.

[0072] According to an implementation of another aspect, this specification also provides a device for classifying samples to be assessed. FIG. 6 is a schematic block diagram illustrating a classification device, according to one implementation. As shown in FIG. 6, a classification device 60 includes: a sample acquisition unit 61, configured to obtain a sample T to be assessed and sample feature Ft of sample T to be assessed; a selection unit 62, configured to select the first quantity N of example samples from a classification sample database; a first acquisition unit 63, configured to obtain feature similarity SIMi between sample T to be assessed and each example sample i of the N example samples, where the feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i; a second acquisition unit 64, configured to obtain sample quality Qi of each example sample i, where sample quality Qi corresponds to such a similarity threshold that historical assessment samples whose feature similarities to the example sample i exceed the similarity threshold are determined in a certain proportion as falling in the category of the classification sample database; a processing unit 65, configured to determine comprehensive similarities Si between sample T to be assessed and each example sample i based on at least difference value ri between feature similarity SIMi and sample quality Qi; and a classification unit 66, configured to determine, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.

[0073] In an implementation, the selection unit 62 includes a calculation subunit (not shown), configured to calculate, based on sample feature Ft of sample T to be assessed and the sample features of the second quantity M of example samples in the classification sample database, feature similarities between each example sample of the second quantity M of example samples and sample T to be assessed, where the second quantity M is greater than the first quantity N; and a selection subunit, configured to select the first quantity N of example samples from the second quantity M of example samples based on feature similarities between each second quantity M of example samples and the sample to be assessed.

[0074] In an implementation, the selection subunit is configured to select, from the second quantity M of example samples, the first quantity N of example samples with the highest feature similarities to sample T to be assessed.

[0075] According to one implementation, the selection unit 62 is configured to select the first quantity N of example samples from the classification sample database based on the sorting result of the sample quality of each example sample in the classification sample database.

[0076] In an implementation, feature similarity SIMi is determined by normalizing the distance between sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.

[0077] According to one implementation, the processing unit 65 is configured to determine comprehensive similarity Si as Si = a + b ri c, where a + b = 1, and c is a coefficient associated with sample quality Qi.

[0078] In an implementation, in the case of ri > = 0, c = 1/(1-Qi) and in the case of ri < 0, c = 1/Qi.

[0079] According to one implementation, the classification unit 66 is configured to determine, based on comprehensive similarities Si between sample T to be assessed and each example sample i, total similarity scores of the sample to be assessed and to determine, based on the total similarity score, whether sample T to be assessed falls in the category of the classification sample database.

[0080] In an implementation, the classification unit 66 is further configured to: if at least one ri > = 0, determine the total similarity score as the maximum value among comprehensive similarities Si between sample T to be assessed and each example sample i; or otherwise, determine the total similarity score as the minimum value among comprehensive similarities Si between sample T to be assessed and each example sample i.

[0081] In an implementation, the classification unit 66 is configured to determine the total similarity score as the average score of the comprehensive similarities Si between sample T to be assessed and each example sample i.

[0082] According to the device in the previous implementation, the feature similarity score of the sample to be assessed and the sample quality of the example sample can be comprehensively considered in determining the comprehensive similarity between the sample to be assessed and the example sample. Therefore, the adverse impact of varied sample quality on the assessment results is reduced or avoided.

[0083] According to another implementation, a computer readable storage medium is also provided, where the computer readable storage medium stores a computer program, and when the computer program is executed in a computer, the computer is enabled to perform the methods described with reference to FIG. 2 to FIG. 5.

[0084] According to another implementation, a computing device is further provided, including a memory and a processor, where the memory stores executable code, and when the processor executes the executable code, the methods described with reference to FIG. 2 to FIG. 5 are implemented.

[0085] A person skilled in the field should be aware that, in one or more of the previous examples, the functions described in the present disclosure can be implemented in hardware, software, firmware, or any combination of them. When these functions are implemented by software, the functions can be stored in a computer readable medium or transmitted as one or more instructions or code on the computer readable medium.

[0086] The specific implementations further describe the object, technical solutions and beneficial effects of the present disclosure. It should be understood that the previous descriptions are merely specific implementations of the present disclosure and are not intended to limit the protection scope of the present disclosure. Any modification, equivalent replacement and improvement made on the basis of the technical solution of the present disclosure shall fall within the protection scope of the present disclosure.


Claims

1. A method for classifying samples to be assessed, comprising:

obtaining sample T to be assessed and sample feature Ft of sample T to be assessed;

selecting the first quantity N of example samples from a classification sample database;

obtaining feature similarity SIMi between sample T to be assessed and each of the N example samples i, wherein feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i;

obtaining sample quality Qi of each example sample i;

determining comprehensive similarities Si between sample T to be assessed and each example sample i based on at least a difference values ri between feature similarity SIMi and sample quality Qi;

determining, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.


 
2. The method according to claim 1, wherein the selecting the first quantity N of example samples from a classification samples comprises:

calculating feature similarities between sample T to be assessed and each of the second quantity M of example samples based on sample feature Ft of sample T to be assessed and sample features of the second quantity M of example samples in the classification sample database, wherein the second quantity M is greater than the first quantity N; and

selecting the first quantity N of example samples from the second quantity M of example samples based on feature similarity between the sample to be assessed and each of the second quantity M of example samples.


 
3. The method according to claim 2, wherein the selecting the first quantity N of example samples from the second quantity M of example samples comprises: selecting, from the second quantity M of example samples, the first quantity N of example samples with the highest feature similarities to sample T to be assessed.
 
4. The method according to claim 1, wherein the selecting the first quantity N of example samples from a classification sample database comprises: selecting the first quantity N of example samples from the classification sample database based on the sorting result of sample quality of each sample in the classification sample database.
 
5. The method according to claim 1, wherein feature similarity SIMi is determined by normalizing the distance between sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.
 
6. The method according to claim 1, wherein the determining comprehensive similarities Si between sample T to be assessed and each example sample i comprises: determining comprehensive similarity Si as Si = a + b ri c, wherein a + b = 1 and c is a coefficient associated with sample quality Qi.
 
7. The method according to claim 6, wherein in the case of ri > = 0, c = 1/(1-Qi) and in the case of ri < 0, c = 1/Qi.
 
8. The method according to claim 1, wherein the determining whether sample T to be assessed falls in the category of the classification sample database based on comprehensive similarity Si comprises:

determining, based on comprehensive similarities Si between sample T to be assessed and each example sample i, a total similarity score of the sample to be assessed; and

determining, based on the total similarity score, whether sample T to be assessed falls in the category of the classification sample database.


 
9. The method according to claim 8, wherein the determining the total similarity score of the sample to be assessed comprises:

if at least one ri > = 0, determining the total similarity score as the maximum value among comprehensive similarities Si between sample T to be assessed and each example sample i; or

otherwise, determining the total similarity score as the minimum value among comprehensive similarities Si between sample T to be assessed and each example sample i.


 
10. The method according to claim 8, wherein the determining a total similarity score of the sample to be assessed comprises: determining the total similarity score as an average value of comprehensive similarities Si between the sample to be assessed and each example sample i.
 
11. A device for classifying samples to be assessed, comprising:

a sample acquisition unit, configured to obtain sample T to be assessed and sample feature Ft of sample T to be assessed;

a selection unit, configured to select the first quantity N of example samples from a classification sample database;

a first acquisition unit, configured to obtain feature similarity SIMi between sample T to be assessed and each of the N example samples i, wherein feature similarity SIMi is determined based on sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i;

a second acquisition unit, configured to obtain sample quality Qi of each example sample i, wherein sample quality Qi corresponds to such a similarity threshold that a sample to be assessed whose feature similarity to the example sample i exceeds the similarity threshold is determined as falling in the category of classification sample database;

a processing unit, configured to determine comprehensive similarities Si between sample T to be assessed and each example sample i based on at least a difference value ri between feature similarity SIMi and sample quality Qi; and

a classification unit, configured to determine, based on comprehensive similarity Si, whether sample T to be assessed falls in the category of the classification sample database.


 
12. The device according to claim 11, wherein the selection unit comprises:

a calculation subunit, configured to calculate, based on sample feature Ft of sample T to be assessed and sample features of the second quantity M of example samples, feature similarities between each of the second quantity M of example samples and sample T to be assessed, wherein the second quantity M is greater than the second quantity N.

a selection subunit, configured to select the first quantity N of example samples from the second quantity M of example samples based on the feature similarities between the sample to be assessed and each of the second quantity M of example samples.


 
13. The device according to claim 12, wherein the selection subunit is configured to select, from the second quantity M of example samples, the first quantity N of example samples with the highest feature similarities to sample T to be assessed.
 
14. The device according to claim 11, wherein the selection unit is configured to select the first quantity N of example samples from the classification sample database based on the sorting result of the sample quality of each sample in the classification sample database.
 
15. The device according to claim 11, wherein feature similarity SIMi is determined by normalizing the distance between sample feature Ft of sample T to be assessed and sample feature Fi of each example sample i.
 
16. The device according to claim 11, wherein the processing unit is configured to determine comprehensive similarity Si as Si = a + b ri c, wherein a + b = 1, and c is a coefficient associated with sample quality Qi.
 
17. The device according to claim 16, wherein in the case of ri > = 0, c = 1/(1-Qi) and in the case of ri < 0, c = 1/Qi.
 
18. The device according to claim 11, wherein the classification unit is configured to:

determine, based on comprehensive similarities Si between sample T to be assessed and each example sample i, a total similarity score of the sample to be assessed; and

determine, based on the total similarity score, whether sample T to be assessed falls in the category of the classification sample database.


 
19. The device according to claim 18, wherein the classification unit is configured to:

if at least one ri > = 0, determine the total similarity score as the maximum value among comprehensive similarities Si between sample T to be assessed and each example sample i; or

otherwise, determine the total similarity score as the minimum value among comprehensive similarities Si between sample T to be assessed and each example sample i.


 
20. The device according to claim 18, wherein the classification unit is configured to determine the total similarity score as the average value of comprehensive similarities Si of sample T to be assessed and each example sample i.
 
21. A computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed in a computer, the computer is enabled to perform the method according to any one of claims 1 to 10.
 
22. A computing device, comprising a memory and a processor, wherein the memory stores executable code, and when executing the executable code, the processor implements the method according to any one of claims 1 to 10.
 




Drawing