CN116739846A

CN116739846A - Trademark risk investigation method, system and device based on big data and storage medium

Info

Publication number: CN116739846A
Application number: CN202310483908.1A
Authority: CN
Inventors: 盛锐; 李海龙; 李家勇; 陆鑫鑫; 崔显龙
Original assignee: Hefei Wanzhen Technology Co ltd
Current assignee: Hefei Wanzhen Technology Co ltd
Priority date: 2023-04-27
Filing date: 2023-04-27
Publication date: 2023-09-12

Abstract

The embodiment of the specification provides a trademark risk investigation method, a system, a device and a storage medium based on big data, wherein the method is realized based on a trademark risk investigation system, and comprises the following steps: acquiring a first data sample, registration classification and international classification of a target trademark; determining at least one pre-estimated audit class based on the registration class and the international class; acquiring at least one estimated examining and classifying reference trademark data based on the big data; determining target feature data of a target trademark through a machine learning model based on the first data sample; matching in the reference trademark data based on the target characteristic data, and determining a similar risk coefficient of the target trademark; and determining the violation risk coefficient of the target trademark by a judging database based on the target characteristic data, wherein the violation risk coefficient at least comprises a form violation risk coefficient.

Description

Trademark risk investigation method, system and device based on big data and storage medium

Technical Field

The present disclosure relates to the field of trademark risk investigation, and in particular, to a trademark risk investigation method, system, device and storage medium based on big data.

Background

The risks existing in the current registered trademark are mainly: risk of query blindness, risk of brand violation, risk of brand duplication/approximation, etc. The various risk factors are carefully considered when registering the trademark, so that irrecoverable losses are avoided.

Aiming at how to analyze and examine trademark risks, a trademark monitoring analysis system is disclosed in the prior art CN110288264a, a trademark registered by the applicant is queried from a data source, the query result is subjected to multidimensional statistics such as legal status of the trademark, trademark application time, trademark application name, trademark registration category, trademark registration address, trademark agency and the like to obtain statistical results of each dimension, and risk early warning analysis is performed on the statistical results. However, this application uses only a database to analyze and screen the risk of the trademark, which is time-consuming and less reliable. Another prior art CN106844551B discloses a method and a system for automatically analyzing the success rate of trademark application based on artificial intelligence, which comprises comparing the name, graphic image or sound file of the trademark to be applied with a trademark inhibition information base based on artificial intelligence, and judging the approximation degree of the content specified by the inhibition clauses of the trademark to be applied and the trademark application; based on artificial intelligence, the name, trademark graph and picture or sound file of the trademark to be applied are compared with the continuous trademark information library, and the similarity between the trademark to be applied and the existing continuous trademark is searched and judged. However, the application only judges the similar risk and the illegal risk based on the name of the trademark to be applied, the trademark graphic picture or the sound file, and the reliability is low.

In view of this, the present specification provides a trademark risk investigation method, system, device and storage medium based on big data, which can improve trademark risk investigation efficiency, reduce labor cost and time cost, improve risk investigation effectiveness and reliability, and can implement intelligent trademark modification.

Disclosure of Invention

One of embodiments of the present disclosure provides a trademark risk checking method based on big data, where the method is implemented based on a trademark risk checking system, and the method includes: acquiring a first data sample, registration classification and international classification of a target trademark; determining at least one pre-estimated audit class based on the registration class and the international class; acquiring at least one estimated examining and classifying reference trademark data based on the big data; determining target feature data of a target trademark through a machine learning model based on the first data sample; matching in the reference trademark data based on the target characteristic data, and determining a similar risk coefficient of the target trademark; and determining the violation risk coefficient of the target trademark by a judging database based on the target characteristic data, wherein the violation risk coefficient at least comprises a form violation risk coefficient.

One of the embodiments of the present specification provides a trademark risk checking system based on big data, which is characterized in that the system includes: the first acquisition module is used for acquiring a first data sample, registration classification and international classification of the target trademark; a first determining module for determining at least one pre-estimated review category based on the registration category and the international category; the second acquisition module is used for acquiring at least one piece of estimated checking classified reference trademark data based on the big data; the second determining module is used for determining target feature data of a target trademark through a machine learning model based on the first data sample; matching in the reference trademark data based on the target characteristic data, and determining a similar risk coefficient of the target trademark; and determining the violation risk coefficient of the target trademark by a judging database based on the target characteristic data, wherein the violation risk coefficient at least comprises a form violation risk coefficient.

One of the embodiments of the present disclosure provides a trademark risk checking device based on big data, the device including at least one processor and at least one memory; at least one memory for storing computer instructions; at least one processor is configured to execute at least some of the computer instructions to implement the big data based brand risk screening method of any of the embodiments of the present specification.

One of the embodiments of the present disclosure provides a computer-readable storage medium storing computer instructions, where when the computer reads the computer instructions in the storage medium, the computer executes the trademark risk profile method based on big data according to any one of the embodiments of the present disclosure.

Drawings

The present specification will be further elucidated by way of example embodiments, which will be described in detail by means of the accompanying drawings. The embodiments are not limiting, in which like numerals represent like structures, wherein:

FIG. 1 is an exemplary block diagram of a big data based brand risk screening system, according to some embodiments of the present specification;

FIG. 2 is an exemplary flow chart of a big data based trademark risk screening method in accordance with some embodiments of the present description;

FIG. 3 is an exemplary schematic diagram illustrating the determination of overall morphology-like risk factors and partial morphology-like risk factors according to some embodiments of the present description;

FIG. 4 is an exemplary flow chart for determining morphological similarity risk coefficients according to some embodiments of the present description;

FIG. 5 is an exemplary diagram illustrating determining semantically similar risk factors according to some embodiments of the present description.

Detailed Description

In order to more clearly illustrate the technical solutions of the embodiments of the present specification, the drawings that are required to be used in the description of the embodiments will be briefly described below. It is apparent that the drawings in the following description are only some examples or embodiments of the present specification, and it is possible for those of ordinary skill in the art to apply the present specification to other similar situations according to the drawings without inventive effort. Unless otherwise apparent from the context of the language or otherwise specified, like reference numerals in the figures refer to like structures or operations.

It will be appreciated that "system," "apparatus," "unit" and/or "module" as used herein is one method for distinguishing between different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

As used in this specification and the claims, the terms "a," "an," "the," and/or "the" are not specific to a singular, but may include a plurality, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus.

A flowchart is used in this specification to describe the operations performed by the system according to embodiments of the present specification. It should be appreciated that the preceding or following operations are not necessarily performed in order precisely. Rather, the steps may be processed in reverse order or simultaneously. Also, other operations may be added to or removed from these processes.

Fig. 1 is an exemplary block diagram of a big data based brand risk screening system, according to some embodiments of the present specification.

As shown in fig. 1, the big data based trademark risk screening system 100 may include a first acquisition module 110, a first determination module 120, a second acquisition module 130, and a second determination module 140.

In some embodiments, the first acquisition module 110 may be configured to acquire a first data sample of the target trademark, a registration classification, and an international classification.

In some embodiments, the first determination module 120 may be configured to determine at least one pre-estimated audit category based on the registration category and the international category.

In some embodiments, the second obtaining module 130 may be configured to obtain the reference trademark data of the at least one pre-estimated review category based on the big data.

In some embodiments, the second determining module 140 may be configured to determine, based on the first data sample, target feature data of the target trademark through a machine learning model; matching in the reference trademark data based on the target characteristic data, and determining a similar risk coefficient of the target trademark; and determining a violation risk coefficient of the target trademark by determining a database based on the target feature data, wherein the violation risk coefficient at least comprises a form violation risk coefficient.

In some embodiments, the similar risk factor comprises a morphological similar risk factor, the target feature data comprises a first morphological feature of the target trademark, and the second determination module is further configured to: determining a first morphological feature through a morphological feature extraction model based on the first data sample, wherein the morphological feature extraction model is a machine learning model; and matching in the reference trademark data based on the first morphological characteristics, and determining morphological similarity risk coefficients.

In some embodiments, the similarity risk coefficient comprises a semantic similarity risk coefficient, the target feature data comprises a semantic feature of the target trademark, and the second determination module is further configured to: determining semantic features through a semantic feature determining model based on the first data sample, wherein the semantic feature determining model is a machine learning model; and matching in the reference trademark data based on the semantic features, and determining the semantic similarity risk coefficient.

In some embodiments, big data based brand risk screening system 100 may also include a recommendation module (not shown in fig. 1). In some embodiments, the recommendation module may be configured to recommend at least one preferred modification to the user based on the violation risk factors and similar risk factors.

For more on the first acquisition module 110, the first determination module 120, the second acquisition module 130, the second determination module 140, and the recommendation module, see the related description below.

It should be noted that the above description of the trademark risk screening system based on big data and the modules thereof is for convenience of description only, and does not limit the present specification to the scope of the illustrated embodiments. It will be appreciated by those skilled in the art that, given the principles of the system, various modules may be combined arbitrarily or a subsystem may be constructed in connection with other modules without departing from such principles. In some embodiments, the first acquisition module 110, the first determination module 120, the second acquisition module 130, and the second determination module 140 disclosed in fig. 1 may be different modules in one system, or may be one module to implement the functions of two or more modules. For example, each module may share one memory module, or each module may have a respective memory module. Such variations are within the scope of the present description.

Fig. 2 is an exemplary flow chart of a trademark risk screening method, according to some embodiments of the present description. In some embodiments, the process 200 may be performed by the brand risk screening system 100. As shown in fig. 2, the process 200 includes the steps of:

step 210, obtaining a first data sample of a target trademark, a registered category, and an international category.

The target trademark may refer to a trademark of risk to be examined.

The first data sample may refer to data related to a target trademark. The data samples may include pictures, text, audio, and the like.

Registration classification may refer to the type of target trademark. Registration classifications may include two-dimensional image branding (e.g., text, two-dimensional graphics, letters, numbers, etc.), three-dimensional logos, color combinations, and the like.

International classification may refer to the belonging classification of a given use commodity of a target trademark. The designation of the commodity may refer to a commodity or service using a target trademark, or the like.

The ways of obtaining the first data sample, registering the classification, and international classification may include user input, or selecting from existing entity data.

At step 220, at least one predictive audit category is determined based on the registration category and the international category.

The audit category may refer to the category to which the brand was attributed at the time of audit. The pre-estimated review category may refer to a review category to which the target trademark may belong.

In some embodiments, the first determination module 120 may determine at least one pre-estimated review category based on the registration category and the international category according to a preset table. The preset table may store a plurality of registration classifications, a plurality of international classifications, and a plurality of correspondence relationships of estimated review classifications. The preset table may be determined based on historical experience.

Step 230, obtaining at least one estimated and inspected classified reference trademark data based on the big data.

The reference trademark data may refer to a database to which the reference trademark corresponds. The reference trademark refers to a registered trademark belonging to the estimated review class. The reference trademark data may include a data sample of the reference trademark, registered classification, international classification, and the like. In some embodiments, the reference trademark data may further include reference trademark data, which includes content and is obtained in a similar manner to the target trademark data, as will be described more below.

In some embodiments, the second obtaining module 130 may use a trademark database/trademark knowledge graph constructed by big data, etc., wherein a trademark belonging to the estimated review class is used as a reference trademark, and a data sample, a registered class, an international class, etc. of the reference trademark are used as reference trademark data.

In step 240, target feature data of the target trademark is determined by the machine learning model based on the first data sample.

The target feature data may refer to data related to features of the target trademark. In some embodiments, the target feature data may include a first morphological feature, a semantic feature, etc. of the target trademark.

The first morphological feature may refer to a feature related to the morphology of the target trademark. For example, morphological features may include the type of text in a trademark, the manner in which the text is arranged, the angle of arrangement, the shape of the graphic, and the like.

In some embodiments, the first morphological feature may include an overall morphological feature, a partial morphological feature of the target trademark.

The overall morphological feature may refer to a morphological feature associated with the entirety of the target trademark. Partial morphological features may refer to morphological features associated with a portion of a target trademark. For example, the target trademark may be divided into at least one portion, and the morphological feature corresponding to each portion is a partial morphological feature.

Semantic features may refer to features related to the semantics of a target trademark. In some embodiments, the semantic features may include text semantic features, graphics semantic features.

Text semantic features may refer to semantic features of text contained in a target trademark. For example, text semantic features may include meanings or features expressed by text content of words, spellings, harmony of spellings, foreign language translations, and the like. For example, the trademark contains the Chinese character "sailing", and the corresponding text semantic features may include "sailing", "qihang", and the like.

The graphic semantic feature refers to a semantic feature of a graphic contained in the target trademark. For example, the graphical semantic features may include meanings of the graphical representation, and the like. The meaning of the different graphical representations may be the same, i.e. both have the same graphical semantic features. See fig. 5 for further description of text semantic features and graphical semantic features.

In some embodiments, the second determination module 140 may determine the target feature data by processing the first data sample based on a machine learning model in a variety of ways. For example, the first morphological feature and semantic feature in the target feature data may be determined using different machine learning models, respectively, and so forth.

In some embodiments, the second determining module 140 may determine the first morphological feature of the target trademark through a morphological feature extraction model based on the first data sample. See fig. 3 for more on the morphology feature extraction model.

In some embodiments, the second determination module 140 may determine the semantic features of the target trademark by a semantic feature determination model based on the first data sample. See fig. 5 for more on the semantic feature determination model.

Step 250, matching in the reference trademark data based on the target feature data, and determining the similar risk coefficient of the target trademark.

The similar risk factor may refer to a risk index that results from inspection of the target trademark due to similarity of the target trademark to the reference trademark. The similar risk factors may be represented by numerical values or other forms. The higher the similarity risk coefficient, the higher the risk index that represents the examination of the target trademark by the reference trademark similar to the target trademark.

The second determining module 140 may determine the similar risk factors of the target trademark based on matching the target feature data in the reference trademark data in a variety of ways.

In some embodiments, the second determining module 140 may determine, based on the big data technique, an associated feature vector in which a vector distance between the feature vector corresponding to the target feature data (i.e., the target feature vector) and the feature vector corresponding to the reference feature data (i.e., the reference feature vector) is greater than a distance threshold; and determining the similarity risk coefficient of the target trademark based on the preset corresponding relation among the vector distance of the target feature vector and the associated feature vector, the vector distance and the similarity risk coefficient. The preset correspondence between the vector distance and the similarity risk coefficient may be preset based on a priori knowledge or historical experience.

In some embodiments, the similar risk factors may include morphological similar risk factors.

The morphological similarity risk factor may refer to a risk factor brought by the target trademark and the reference trademark being similar in morphological characteristics to the target trademark examination.

The morphological similarity risk factor may be determined in a number of ways. In some embodiments, the second determination module 140 may determine the morphological similarity risk factor based on matching the first morphological feature in the reference trademark data. In some embodiments, the matching means may be: comparing the first morphological characteristics of the target trademark with the reference morphological characteristics of each reference trademark in the reference trademark data one by one, and determining the morphological matching degree of the first morphological characteristics and each reference morphological characteristic; determining a reference morphology feature with morphology matching degree larger than a morphology matching degree threshold as a first associated morphology feature; and carrying out weighted summation on the form matching degree corresponding to one or more first associated form features to obtain the form similarity risk coefficient of the target trademark. The form matching degree can be determined through a vector distance between the first form feature of the target trademark and a feature vector corresponding to the reference form feature. The form matching degree threshold value can be preset. The first weights corresponding to each of the first associated morphology features may be determined in relation to the respective corresponding morphology matching degree. For example, the greater the morphology matching, the greater the corresponding first weight.

In some embodiments, the second determination module 140 may determine the morphological similarity risk factor based on the associated trademark. See fig. 4 for more details regarding this embodiment.

In some embodiments, the morphological similarity risk coefficient may include an overall morphological similarity risk coefficient and a partial morphological similarity risk coefficient.

The overall morphological similarity risk coefficient refers to a risk index brought by the examination of the target trademark by the overall morphological similarity of the target trademark and the reference trademark.

The partial morphological similarity risk coefficient may refer to a risk index caused by that the target trademark is inspected by partial morphological feature similarity of the target trademark and the reference trademark.

The overall morphological similarity risk coefficient may be determined in a variety of ways. For example, the user may perform analysis and judgment according to the target trademark and the reference trademark, so as to determine the overall morphological similarity risk coefficient of the target trademark.

The partial morphological similarity risk factor may be determined in a number of ways. For example, the target trademark may be divided into a plurality of portions, and the user performs analysis and judgment according to the target trademark and the reference trademark, determines sub-morphological similar risk coefficients of each portion, and performs weighted summation on the sub-morphological similar risk coefficients of each portion to obtain the partial morphological similar risk coefficients of the target trademark. The weight of each part may be set manually.

In some embodiments, the second determination module 140 may determine the overall morphology-similar risk coefficient and the partial morphology-similar risk coefficient by a morphology feature extraction model. See fig. 3 for a more description of this embodiment.

In some embodiments, the similarity risk coefficient may also include a semantic similarity risk coefficient.

The semantically similar risk coefficient may refer to a risk index brought about by semantically similar of the target trademark and the reference trademark to the target trademark examination.

The semantically similar risk factor may be determined in a number of ways. In some embodiments, the second determining module 140 may determine the semantic similarity risk coefficient based on matching semantic features of the target trademark in the reference trademark data. The matching method for determining the semantic similarity risk coefficient is similar to the matching method for determining the morphological similarity risk coefficient, and more description is given above, and will not be repeated here.

In some embodiments, the second determination module 140 may determine the semantic similar risk coefficients through a semantic feature determination model. See fig. 5 for a detailed description.

Step 260, determining, based on the target feature data, a violation risk factor of the target trademark by determining a database, the violation risk factor including at least a formal violation risk factor.

The decision database may refer to a database of related content containing inhibition information. The prohibition information may include, among other things, violation of national laws and regulations and possibly damage moral fashion. For example, the decision database may include feature pictures, semantic features, sound files, etc. of prohibition information such as prohibition clause names, prohibition clause contents, prohibition clause interpretations, prohibition clause features, prohibition clauses, etc. As a specific example, the decision database may include some words, phrases, sentences that are prone to adverse effects. The decision database may be obtained from big data. In some embodiments, the decision database may be updated periodically based on the current social hotspot, the attention of the public.

The violation risk factor may refer to a likelihood of violation of the target trademark against the prohibition information. The risk of violation coefficient may be represented by a numerical value or other form. The higher the risk factor of violations of the target trademark, the higher the probability of target trademark violations, and the greater the impact on target trademark passing inspection.

The formal violation risk factor may refer to a likelihood that the morphological feature contained in the target commodity violates the forbidden information.

In some embodiments, the second determining module 140 may determine, based on matching the target feature data in the decision database, a degree of illicit matching between the target feature data and feature pictures corresponding to the respective inhibition information; determining the feature picture with the violation matching degree larger than the violation matching degree threshold as a violation feature picture; and determining the violation risk coefficient of the target trademark based on the violation matching degree of the one or more violation feature pictures. Wherein, the threshold of the matching degree against rules can be preset by people or a system.

The rule-breaking matching degree can refer to the similarity between the target trademark and the characteristic picture corresponding to certain forbidden information in the judging database. The higher the matching degree of the violation is, the higher the similarity between the target trademark and the characteristic picture corresponding to certain forbidden information in the judging database is.

In some embodiments, the violation match may be determined based on a matching model, and the target feature data input as the target trademark and the decision database are output as the violation match of one or more violation feature pictures. The matching model is a machine learning model. Such as a deep neural network model. The matching model can be obtained through supervision training based on a large number of first training samples with first labels, wherein the first training samples comprise historical feature data corresponding to historical examination trademarks and historical violation feature pictures, and the first labels are historical violation matching degrees corresponding to the historical violation feature pictures. The first training sample and the first tag may be obtained based on historical data.

In some embodiments, the violation risk factor for the target trademark may be determined in a number of ways based on the violation match of one or more violation feature pictures. For example, the violation risk factor may be positively correlated to the violation match of the violation feature picture with the highest violation match. As another example, the offending risk coefficient may be positively correlated to the number of offending feature pictures. The specific determination mode can be set according to actual requirements. For example, when the actual demand is the highest offending matching degree of the more important offending feature picture, then the offending risk coefficient may be positively correlated to the offending matching degree of the offending feature picture with the highest offending matching degree.

In some embodiments, the second determining module 140 may determine the violation risk factor of the target trademark by weighted summation based on the violation match of each violation feature picture and its corresponding second weight. Wherein the second weight may be positively correlated to the corresponding offending matching degree of the offending feature picture, which may be set based on a priori knowledge or historical experience.

In some embodiments, the second determination module 140 may present the determined one or more offending feature pictures to the user. By displaying the violation feature pictures to the user, the user can be facilitated to determine whether the target trademark is similar to some forbidden information.

In some embodiments, the second determination module 140 may determine the formal violation risk factor based on matching in the first morphological feature decision database of the target trademark. Wherein, the manner of determining the form of the violation risk factor is similar to that of determining the violation risk factor, and more description can refer to the relevant content.

In some embodiments, the violation risk factors may also include semantic violation risk factors.

The semantic violation risk factor may refer to a likelihood that the semantic feature contained in the target commodity violates the prohibition information.

In some embodiments, the second determination module 140 may determine text semantic features and graphic semantic features of the target trademark based on the first data sample; and determining a semantic violation risk coefficient through a judgment database based on the text semantic features and the graphic semantic features. For more description of text semantic features and graphic semantic features see step 240, FIG. 5 and their associated descriptions.

In some embodiments, the semantic features including the inhibition information in the decision database may include text semantic features and graphic semantic features, and the second determining module 140 may respectively match in the decision database based on the text semantic features and the graphic semantic features of the target trademark, to determine a text semantic violation risk coefficient and a graphic semantic violation risk coefficient; and determining the semantic violation risk coefficient based on the text semantic violation risk coefficient and the graphic semantic violation risk coefficient. For example, the text semantic violation risk coefficient and the graphic semantic violation risk coefficient may be weighted and summed to determine a semantic violation risk coefficient, the weights of which may be preset by the system or by human beings. The method for determining the text semantic violation risk coefficient and the graphic semantic violation risk coefficient is similar to the method for determining the violation risk coefficient, and more description can refer to the relevant content.

In some embodiments of the present disclosure, by determining a semantic violation risk factor, a user may be alerted to a violation risk that exists in the semantics of the target trademark, so that the user determines a modification policy for the target trademark.

In some embodiments, the violation risk factor may also include a significance violation risk factor.

Saliency may refer to the extent to which a target commodity is susceptible to distinguishing other trademarks. For example, a significant portion of the trademark may include the borders, underlines, etc. of the trademark.

The saliency violation risk coefficient may refer to a risk index of the saliency of the target trademark on the target trademark review.

In some embodiments, the saliency violation risk factor is determined based on saliency of the corresponding partial morphology features of the respective portions of the target trademark. For example, the saliency violation risk factor may be inversely related to the sum of saliency of the respective partial morphological features of the target trademark. When the saliency is lower than the saliency threshold, the target trademark can be considered to have no saliency, and the corresponding saliency violation risk coefficient is higher. The saliency threshold may be manually set.

In some embodiments, the correspondence between saliency and the saliency violation risk factor may be predetermined based on prior knowledge or historical experience, etc.; and determining the saliency violation risk coefficient of the target trademark based on the saliency of the target trademark and the corresponding relation.

In some embodiments, the significance may be determined by the duty cycle of the significant portion of the trademark. In some embodiments, the saliency may be determined by a saliency determination model. See fig. 3 for more description of a saliency determination model.

In trademark review, if the target trademark does not have a significant feature, it will not pass trademark review. By determining the significance of the current target trademark through some embodiments of the present description, the possibility that the target trademark passes trademark auditing can be reflected to a certain extent.

In some embodiments, the big data based brand risk screening system may further comprise a recommendation module operable to recommend at least one preferred modification to the user based on the violation risk factors and similar risk factors.

The modification scheme may refer to a scheme of modifying the target trademark. The preferred modification may refer to a modification with a higher rate of audit pass. The risk factor of the violations of the target trademark can be reduced after the target trademark is modified according to the preferred modification scheme.

In some embodiments, the recommendation module may determine a plurality of modification schemes (e.g., a plurality of modification schemes may be obtained after modifying at least one of the target trademark with text, pinyin, letters, numbers, etc.) according to at least one of the similarity risk coefficient, the morphological similarity risk coefficient, the semantic similarity risk coefficient, the violation risk coefficient, the formal violation risk coefficient, and the semantic violation risk coefficient, re-evaluate the violation risk coefficients and/or the similarity risk coefficients of the respective modified trademark, and determine a modification scheme below the violation risk coefficients and/or the similarity risk coefficients below the corresponding thresholds as the preferred modification scheme.

In some embodiments, the recommendation module may determine the at least one preferred modification in response to the similar risk coefficient being greater than the similar risk threshold and/or the offending risk coefficient being greater than the offending risk threshold.

The similarity risk threshold may refer to a maximum allowable value of the similarity of the target trademark to the reference trademark. The similarity risk threshold may be determined manually. When the similarity of the target trademark with the reference trademark exceeds the similarity risk threshold value, it means that the target trademark may be considered to be the same as or similar to the registered trademark, and is not allowed to be granted trademark rights.

The violation risk threshold may refer to a maximum allowable value of the similarity of the target trademark with the prohibition information. Details of the similarity risk threshold may be referred to for details.

In some embodiments, the recommendation module may modify the direction by determining a priority value for at least one intelligence; generating a plurality of groups of candidate modification schemes based on the priority value of at least one intelligent modification direction; evaluating trademark violations and similar risk reductions for each set of candidate modifications; at least one preferred modification is determined based on the brand violations and similar risk reductions for each set of candidate modifications.

The intelligent modification direction may refer to a modification direction to the target trademark. In some embodiments, the intelligent modification direction may include a base element modification and/or a heuristic modification.

In some embodiments, the modification of the base element may include modifying the sub-direction of a single color modification within the wholly or partially enclosed region, a modification to the location information of an independent sub-portion, and the like. The separate subsection may be a separate subsection of the target trademark that is not connected to other sections of the target trademark. The modification of the position information of the independent sub-parts may include modifying the sub-directions by translation, rotation, scaling, flipping left and right, etc.

In some embodiments, heuristic modifications may include modifications that are non-standard, and are not predictable. For example, the heuristic modification may change a certain rectangular shape in the target trademark to a circular shape. For example, heuristic modifications may generate new graphics, text elements, etc. in the target trademark. As another example, heuristic modifications may replace text in the target trademark with other text of the same type, or similar semantics, or the like.

The priority value may be used to indicate the degree of tendency to take preference for a certain intelligent modification direction when modifying the target trademark. For the intelligent modification directions with larger priority values, in the generated multiple groups of candidate modification schemes, the more schemes containing the intelligent modification directions are, and the more modification sub-directions of the intelligent modification directions in each candidate modification scheme are. The basic element modifies the corresponding priority value to be a first priority value, and the heuristic modification corresponding priority value to be a second priority value.

In some embodiments, the priority value of the intelligent modification direction is related to a morphological similarity risk coefficient and a semantic similarity risk coefficient of the target trademark. Accordingly, a priority value of the intelligent modification direction may be determined based on the morphological similarity risk coefficient and the semantic similarity risk coefficient.

For example, when the semantic similarity risk coefficient of the target trademark is large, the semantic similarity risk coefficient cannot be reduced only by modifying the basic element without changing the semantic, and heuristic modification of the target trademark should be prioritized. When the morphological similarity risk coefficient of the target trademark is large, both the basic element modification and the heuristic modification can be performed, but from the perspective of saving resources (the heuristic modification needs to be used for the model), the basic element modification can be preferentially used.

In some embodiments of the present disclosure, determining the priority value of the intelligent modification direction based on the morphological similarity risk coefficient and the semantic similarity risk coefficient may improve efficiency of determining the priority value, and provide a basis for determining the modification direction of the target trademark.

The candidate modification scheme may refer to a modification scheme that modifies the target trademark in accordance with the intelligent modification direction. The types of the candidate modification schemes may include a scheme of performing basic element modification to the target trademark (hereinafter referred to as an "element modification scheme"), a scheme of performing heuristic modification to the target trademark (hereinafter referred to as a "heuristic modification scheme"), and a scheme of simultaneously performing basic element modification and heuristic modification to the target trademark (hereinafter referred to as a "simultaneous modification scheme").

In some embodiments, the recommendation module may determine a quantitative ratio of element modifications, heuristic modifications, simultaneous modifications among the plurality of sets of candidate modifications based on the priority value of the at least one intelligent modification direction. In some embodiments, the quantitative proportion may be positively correlated to a priority value of the intelligent modification direction. For example, when the first priority value is higher than the second priority value, the number proportion of element modifications is greater than the number proportion of simultaneous modifications, while the number proportion of modifications is greater than the number proportion of heuristic modifications.

In some embodiments, when the basic element of the target trademark is modified, the recommendation module may randomly select a certain number of modification sub-directions to modify for a plurality of times, and determine a certain number of element modification schemes.

In some embodiments, when the target trademark is modified heuristically, the recommendation module may process the target trademark by generating a model to determine a certain number of heuristically modified schemes. The generative model may be a machine learning model, for example, the generative model may be Auto-Encoder or the like.

In some embodiments, the generated model may be obtained by performing a supervised training with a plurality of second training samples and second labels. The second training sample includes a historical trademark, which may be obtained based on historical data. The second label can be a history trademark modified by heuristic, and the second label can be obtained by manual labeling.

In some embodiments, the generative model may also be trained simultaneously in conjunction with other discriminant models, for example, by way of GAN (generated against a network). The discriminant model may be a pre-trained machine learning model. In the training process, the output of the generated model can be used as a false sample, the second label of the second training sample is used as a true sample, the false sample and the true sample are input into the judging model at the same time, and the similarity of the false sample and the true sample is output. And continuously updating parameters of the generated model, and completing the training of the generated model when the similarity between the true sample and the false sample is higher and higher until the discrimination model judges the false sample as the true sample.

According to the method and the device for generating the model, the generating model is trained in a manner of generating the countermeasure network, and the generating model and the judging model are enabled to be stronger in capability through continuous circulation of self-game in the training process, so that the output accuracy of the generating model is improved.

In some embodiments, when it is determined that the basic element modification and the heuristic modification are performed on the target trademark at the same time, the recommendation module may first perform the basic element modification, and then determine the third candidate modification scheme by generating the model based on the modification result. See above for relevant description.

Trademark violations may refer to the degree of violations of the modified target trademark.

In some embodiments, the brand violations of the modified target brand may be determined based on the violations determination model. The violation determination model is a machine learning model, for example, a deep neural network model. In some embodiments, the violation determination model may be obtained through a number of third training samples and third labels for supervised training. The third training sample includes a historical trademark, which may be obtained based on historical data. The third label may be the degree of violation of the historical trademark, and the third label may be manually marked. For example, a historical trademark that the human eye looks like is more offensive may be marked as having a higher trademark violation. For example, if there are more obvious flaws in the trademark such as distortion of lines, double images, overlapping of sub-elements, etc., the trademark violation can be marked as higher.

The similarity risk reduction degree may refer to a degree to which the modified target trademark has reduced similarity to the reference trademark. Illustratively, the similarity risk reduction degree may be determined by the following formula (1).

Where e is the similarity risk reduction, ω is the similarity risk factor for the target trademark, Is a similar risk factor for the modified trademark. See step 250 for more description of determining similar risk factors.

In some embodiments, the recommendation module may determine one or more candidate modifications with brand violations below a violation threshold and similar risk reductions above a reduction threshold as one or more preferred modifications. Wherein the violation threshold and the reduction threshold may be preset by the system or by human.

In some embodiments of the present disclosure, the candidate modification schemes are determined based on the priority values, so that a more suitable candidate modification scheme can be effectively determined. The optimal modification scheme is determined according to the trademark violation degree and the similarity risk reduction degree, so that the efficiency and the accuracy of determining the optimal modification scheme can be improved.

In some embodiments, the recommendation module may also evaluate a degree of change in the salient features for each set of candidate modifications; at least one preferred modification is determined based on the degree of change in the salient features, the degree of brand violation, and the degree of similar risk reduction.

The degree of change in the salient feature may refer to a degree of change in the saliency of the modified target trademark. The higher the degree of change of the salient features, the greater the degree of change of the salient features before and after modification.

In some embodiments, the degree of change in the salient features may be determined based on a change in the degree of change in the part of the morphological feature of the modified target trademark. Illustratively, the salient feature variation degree may be determined by formula (2).

Where ρ is the degree of change of the salient feature, τ is the degree of saliency of the target trademark, and δ is the degree of saliency of the modified target trademark. See fig. 3 for more description of determining saliency.

In some embodiments, the recommendation module may determine one or more candidate modifications having a significant feature change below a change threshold, a brand violation below a violation threshold, and a similar risk reduction above a reduction threshold as one or more preferred modifications. Wherein the change threshold may be preset by the system or by human.

In some embodiments of the present disclosure, determining the preferred modification based on the degree of change in the salient features may effectively and purposefully improve the uniqueness of the trademark. For example, a user may wish the text to have a high degree of saliency, and the corresponding intelligent modification direction may not be biased toward shifting the saliency into the graphic.

According to some embodiments of the specification, the estimated examination classification is determined according to the related information of the target trademark, so that the examination scope can be effectively reduced, the similarity of the target trademark and the reference trademark can be compared in a targeted manner, the risk of the applied trademark is further determined, and the accuracy of risk determination can be improved. Further, according to the similarity between the target trademark and the reference trademark, a proper preferred modification scheme can be quickly determined and recommended to the user, and the application passing rate of the trademark is effectively improved.

FIG. 3 is an exemplary schematic diagram illustrating the determination of overall morphology-like risk factors and partial morphology-like risk factors according to some embodiments of the present description.

In some embodiments, the second determination module 140 may determine the first morphology feature by a morphology feature extraction model based on the first data sample; and matching in the reference trademark data based on the first morphological characteristics, and determining morphological similarity risk coefficients. See fig. 2 for a more description of determining morphological similarity risk coefficients.

The morphology feature extraction model may be a machine learning model, for example, a convolutional neural network model.

In some embodiments, the input of the morphology feature extraction model may include a first data sample of the target trademark and the output may be a first morphology feature. See fig. 2 for a more description of the first morphology feature.

In some embodiments, the morphology feature extraction model may be trained from a plurality of third training samples with third labels. In some embodiments, the third training sample may be a first data sample of a plurality of historical trademarks, and the third training sample may be obtained based on the historical data. The third label may be a historical first morphology feature corresponding to a historical trademark. The third label may be manually labeled.

In some embodiments of the present disclosure, the accuracy and efficiency of determining the morphological similarity risk coefficient may be improved by determining the first morphological feature through the morphological feature extraction model and further determining the morphological similarity risk coefficient.

In some embodiments, the morphological similarity risk coefficient may include an overall morphological similarity risk coefficient and at least one partial morphological similarity risk coefficient. See fig. 2 for a more explanation of overall morphology-like risk factors and partial morphology-like risk factors.

In some embodiments, the overall morphology-like risk factor and the partial morphology-like risk factor may be determined by analyzing the first data sample of the target trademark.

In some embodiments, the overall morphology-like risk coefficient and the partial morphology-like risk coefficient may be determined by a morphology feature extraction model.

Some of the examples described below may be understood with reference to fig. 3, which is merely illustrative of some of the embodiments thereof and not limiting of the embodiments.

In some embodiments, morphology feature extraction model 320 includes a first feature extraction layer 321 and a second feature extraction layer 322.

In some embodiments, the overall morphology similarity risk factor 370 is determined by: determining, based on the first data sample 310, overall morphological features 350 of the target trademark through the first feature extraction layer 321; the overall morphology similarity risk factor 370 is determined based on matching the overall morphology features 350 of the target trademark in the reference trademark data 360.

In some embodiments, the determination of the partial morphology-like risk factor 380 includes: determining at least one partial morphological feature 340 of the target trademark by the second feature extraction layer 322 based on the first data sample 310; the partial morphological similarity risk factor 380 is determined based on the matching of the at least one partial morphological feature 340 in the reference trademark data 360.

The first feature extraction layer 321 may be a machine learning model, for example, the first feature extraction layer 321 may be a convolutional neural network model.

In some embodiments, the input of the first feature extraction layer 321 may comprise the first data sample 310 and the output may comprise the global morphological feature 350. See fig. 2 for a more description of overall morphology features.

In some embodiments, the global morphological similarity risk factor 370 may be determined based on matching global morphological features 350 in the reference trademark data 360. The manner of determining the overall morphology-like risk factor is similar to that of determining the morphology-like risk factor, and more is described with reference to fig. 2.

The second feature extraction layer 322 may be a machine learning model, for example, the second feature extraction layer 322 may be a convolutional neural network model.

In some embodiments, the input of the second feature extraction layer 322 may comprise the first data sample 310 and the output may comprise at least one partial morphology feature 340.

In some embodiments, the partial morphology features may include sub-region morphology features.

The morphological characteristics of the subareas can refer to morphological characteristics of subareas obtained by cutting the target trademark.

In some embodiments, the second determination module 140 may determine the at least one sub-region morphology feature 334 through the second feature extraction layer 322. In some embodiments, the second feature extraction layer 322 may include a sub-region morphology feature extraction module 330 and the second determination module 140 may determine the at least one sub-region morphology feature 334 through the sub-region morphology feature extraction module 330.

In some embodiments, the sub-region morphology feature extraction module 330 may include a region segmentation layer 331 and a sub-region morphology feature determination layer 333 that are connected to each other.

In some embodiments, the region segmentation layer 331 may determine at least one sub-region image 332, such as sub-region image 1, sub-region images 2, … …, sub-region image n, based on the first data sample 310. At least one sub-region image 332 output by the region division layer 331 may be input to the sub-region morphological feature determination layer 333.

The sub-region morphology feature determination layer 333 may determine at least one sub-region morphology feature 334, such as sub-region morphology feature 1, sub-region morphology features 2, … …, sub-region morphology feature n, based on the at least one sub-region image 332.

In some embodiments, the region segmentation layer 331 may be a machine learning model, such as a yolo model. In some embodiments, the region segmentation layer 331 may be a non-machine learning model. For example, the region dividing layer 331 may be a preset dividing rule, such as equidistant parallel dividing. The sub-region morphology feature determination layer 333 may be a machine learning model, for example, a convolutional neural network.

In some embodiments of the present disclosure, by dividing the target trademark into different subareas, determining morphological features of the subareas according to the subareas, and further determining a partial morphological similarity risk coefficient, the similarity between the target trademark and the reference trademark can be determined from the area level, and further accuracy of the similarity risk coefficient is improved.

In some embodiments, the partial morphology features may include subelement morphology features.

The sub-element morphological feature may refer to a morphological feature of a sub-element contained in the target trademark. The sub-elements may include the target trademark's chinese characters, letters, numbers, sub-graphics, etc.

In some embodiments, the sub-element morphological features may include at least one of a kanji element morphological feature, an alphabetic element morphological feature, a numeric element morphological feature, a graphical element morphological feature.

The morphological features of the chinese character elements may refer to morphological features associated with chinese characters in the target trademark. Morphological features may include at least one of font, chinese, letter, number, graphic, shape, color, size, etc. The letter element morphology feature may refer to morphology features associated with letters in the target trademark. The numerical element morphological feature may refer to a morphological feature associated with a number in a target trademark. The graphic element morphological feature may refer to a morphological feature associated with a graphic in a target trademark.

In some embodiments, the second determination module 140 may determine at least one sub-element morphological feature through the second feature extraction layer 322. For example, the first data sample 310 is input into the second feature extraction layer 322 and output results in at least one sub-element morphological feature 344.

In some embodiments, the second feature extraction layer 322 may include a sub-element morphology feature extraction module 340 and the second determination module 140 may determine at least one sub-element morphology feature 344 through the sub-element morphology feature extraction module 340.

In some embodiments, the sub-element morphology feature extraction module 340 may include an element segmentation layer 341 and a sub-element morphology feature determination layer 343 that are connected to each other.

In some embodiments, the element segmentation layer 341 may determine at least one sub-element image 342, e.g., a kanji element image, a letter element image, … …, a graphic element image, based on the first data sample 310. At least one sub-element image 342 output by the element segmentation layer 341 may be input to a sub-element morphological feature determination layer 343.

The sub-element morphological feature determination layer 343 may determine at least one sub-element morphological feature 344, such as a kanji element morphological feature, a letter element morphological feature, … …, a graphic element morphological feature, based on the at least one sub-element image 342. Each class of subelement images can correspond to a class of subelement morphological features.

In some embodiments, the element segmentation layer 341 may be a machine learning model, e.g., yolo model. In some embodiments, the subelement morphology feature determination layer 343 may be a machine learning model, such as a convolutional neural network.

In some embodiments of the present disclosure, by dividing the target trademark into different elements and determining the morphological characteristics of the sub-elements according to the different elements, and further determining the partial morphological similarity risk coefficient, the similarity between the target trademark and the reference trademark can be determined from the element layer of the wrapping and replacing, and further the accuracy of the similarity risk coefficient is improved.

In some embodiments, for any part of the morphological features, matching may be performed in the reference trademark data based on the part of the morphological features, and the matching similarity between the part of the morphological features and the part of the morphological features of each reference trademark in the reference trademark data is determined; taking the reference trademark with the matching similarity larger than the similarity threshold value as the matching reference trademark; carrying out weighted summation based on the matching similarity of the plurality of matching reference trademarks and the third weight thereof to obtain the matching similarity weighted sum of the partial morphological characteristics as the initial similarity risk coefficient of the partial morphological characteristics; and finally, carrying out weighted summation based on the initial similar risk coefficient of each partial morphological feature and the fourth weight thereof to obtain the partial morphological similar risk coefficient.

The matching similarity may refer to a similarity of a part of morphological features of the target trademark and a part of morphological features of the reference trademark. The matching similarity may be determined based on the vector distance between the two, and for more description, reference may be made to fig. 2 for a description of the determination of morphology matching.

In some embodiments, the third weight may be positively correlated to the matching similarity, which may be set according to actual requirements.

In some embodiments, the fourth weight may be related to a significance of the at least one partial morphology feature in the first data sample. In some embodiments, the saliency is obtained by a saliency determination model.

The saliency determination model may be used to determine a corresponding saliency of at least one partial morphology feature. The saliency determination model may be a machine learning model, e.g., a deep neural network model, etc.

In some embodiments, the input of the saliency determination model is a first data sample of the target trademark and at least one partial morphological feature, and the output is a saliency corresponding to each partial morphological feature. In some embodiments, the input of the saliency determination model may be a first data sample of the target trademark and at least one sub-region image (or sub-region morphological feature), and the saliency corresponding to each sub-region morphological feature is output. In some embodiments, the input of the saliency determination model may be a first data sample of the target trademark and at least one sub-element image (or sub-element morphological feature), and the saliency corresponding to each sub-element morphological feature is output.

In some embodiments, the saliency determination model may be trained from a number of fourth training samples with fourth labels.

In some embodiments, the fourth training sample may be a plurality of historical first data samples of historical trademarks and historical partial morphology features, and the fourth training sample may be obtained based on the historical data. The fourth label of the fourth training sample may have a historical saliency corresponding to the historical partial morphology feature. The fourth label may be manually labeled. When the saliency marking is performed manually, the mark can be performed by referring to trademark examination guidelines. For example, when the trademark has only a common name, a graphic, or a model of the present commodity, it may be noted that the significance thereof is low.

According to some embodiments of the specification, the influence of the more remarkable target trademark part on the morphological similarity risk coefficient can be improved by determining the weight of each part morphological feature through the saliency, and the accuracy of the similarity risk coefficient is improved.

In some embodiments of the present disclosure, determining the overall morphological similarity risk coefficient and the partial morphological similarity risk coefficient, respectively, may improve accuracy of determining the similarity risk coefficient.

Fig. 4 is an exemplary flow chart for determining morphological similarity risk factors according to some embodiments of the present description. In some embodiments, the process 400 may be performed by the second determination module 140. As shown in fig. 4, the process 400 includes the steps of:

And step 410, modifying the basic element of the first data sample to obtain at least one associated trademark.

The related trademark may refer to a trademark obtained by modifying a basic element of the target trademark. For example, the associated trademark may be obtained by changing the red line in the target trademark to a yellow line. See fig. 2 for a more description of the basic element modification.

Step 420, determining the second data sample of the at least one associated trademark as an associated trademark cluster of the target trademark.

The second data sample may refer to at least one data sample of an associated trademark. The second data sample is similar to the first data sample, for more explanation, see the associated description of the first data sample in fig. 2.

The associated trademark cluster may refer to a collection of second data samples of at least one associated trademark. The determination of the associated trademark clusters may refer to the determination of the first data samples.

And step 430, determining a second morphological feature of at least one associated trademark in the associated trademark cluster through the morphological feature extraction model. See fig. 3 for more description of morphology feature extraction models.

The second morphological feature may refer to a morphological feature of the associated trademark. The second morphology feature is similar to the first morphology feature, see fig. 2 for more explanation.

The second morphology feature may be a model by morphology feature extraction. In some embodiments, the input of the morphology feature extraction model may include a second data sample of the associated trademark and the output may include a second morphology feature.

Step 440, determining a morphological similarity risk coefficient based on the first morphological feature of the target trademark and the second morphological feature of the at least one associated trademark matching in the reference trademark data.

In some embodiments, the matching means may be: comparing the first morphological feature of the target trademark, the second morphological feature of the related trademark and the reference morphological feature of each reference trademark in the reference trademark data one by one, determining the first morphological matching degree of the first morphological feature and each reference morphological feature, and determining the second morphological matching degree of the second morphological feature and each reference morphological feature; determining the reference morphological feature with the first morphological matching degree or the second morphological feature larger than the morphological matching degree threshold as a second associated morphological feature; and carrying out weighted summation on the first form matching degree or the second form characteristic corresponding to the one or more second associated form characteristics to obtain the form similarity risk coefficient of the target trademark. The determining manner of the first morphology matching degree and the second morphology matching degree can refer to the determining manner of the morphology matching degree, and more description about the morphology matching degree and the morphology matching degree threshold is shown in fig. 2. The fifth weight corresponding to each second associated morphology feature may be determined in relation to the respective corresponding morphology matching degree. For example, the greater the degree of matching of the first or second shape, the greater the corresponding fifth weight.

According to some embodiments of the specification, the related trademark is obtained through modification of the basic element, and is used as a main body for matching, and the related trademark is matched with the reference trademark data.

In some embodiments, the second determination module 140 may determine the semantic features by a semantic feature determination model based on the first data sample; and matching in the reference trademark data based on the semantic features, and determining the semantic similarity risk coefficient. See fig. 2 for a more description of the first data sample, semantic features, and semantic similarity risk coefficients.

The semantic feature determination model may be a machine learning model, e.g., a deep neural network model, etc.

In some embodiments, the input of the semantic feature determination model may be a first data sample and the output may be a semantic feature.

In some embodiments, the semantic feature determination model may be trained based on a plurality of fifth training samples and fifth tags.

In some embodiments, the fifth training sample is a sample first data sample of the sample trademark, and the fifth label is a semantic feature corresponding to the sample first data sample. Wherein, the sample trademark refers to the trademark of historical examination. The fifth training sample can be obtained from a commercial label database, and the fifth label can be determined by means of manual labeling or automatic labeling.

In some embodiments, the semantic features may include text semantic features as well as graphic semantic features. See fig. 2 for more description. In some embodiments, text semantic features as well as graphics semantic features may be determined by a semantic feature determination model.

Some of the examples described below may be understood with reference to fig. 5, which is merely illustrative of some of the embodiments thereof and not limiting of the embodiments.

In some embodiments, the semantic feature determination model 520 may include a segmentation layer 521, a text semantic determination layer 524, and a graphical semantic determination layer 525.

The segmentation layer 521 may determine the text element image 522 and the graphic element image 523 based on the first sample data 510. In some embodiments, the segmentation layer 521 may be a machine learning model, such as a convolutional neural network model, or the like.

The text element image refers to a segmented image containing only text. For example, the text element image may be an image including at least one text of Chinese characters, pinyin, english, other languages, and the like.

The graphic element image refers to an image that is divided to include only graphics. For example, the graphic element image may be an image including at least one graphic of a circle, an arc, or the like.

The text element image 522 output by the segmentation layer 521 may be input to the text semantic determination layer 524, and the graphic element image 523 output may be input to the graphic semantic determination layer 525.

The text semantic determination layer 524 may determine text semantic features 531 based on the text element images 522. In some embodiments, the text semantic determination layer 524 may be a machine learning model. Such as convolutional neural network models, and the like. See fig. 2 for more description of text semantic features.

The graphic semantic determination layer 525 may determine graphic semantic features 532 based on the graphic element images 523. In some embodiments, the graphical semantic determination layer 525 may be a machine learning model. Such as convolutional neural network models, and the like. See fig. 2 for a more description of graphical semantic features.

In some embodiments of the present disclosure, the first data sample is divided into a text element image and an image element image by the dividing layer, then text semantic features are determined based on the text element image by the text semantic determining layer, and graphic semantic features are determined based on the graphic element image by the graphic semantic determining layer, so that different semantic features in the first data sample are extracted accurately, and a semantic similarity risk coefficient is determined more accurately.

In some embodiments, the segmentation layer, the text semantic determination layer, and the graphical semantic determination layer may be obtained in a joint training. In some embodiments, the sixth training sample of the joint training may be a historical first data sample corresponding to a plurality of historical target brands, and the sixth training sample may be obtained based on the historical data. The sixth label corresponding to the sixth training sample may be a corresponding text semantic feature and a graphic semantic feature. The sixth label may be manually labeled.

An exemplary joint training process includes: inputting historical first data samples corresponding to a plurality of historical target trademarks into an initial segmentation layer to obtain a text element image and a graphic element image which are output by the initial segmentation layer; inputting the text element image output by the initial segmentation layer into an initial text semantic determining layer to obtain text semantic features output by the initial text semantic determining layer; inputting the graphic element image output by the initial segmentation layer into an initial graphic semantic determining layer to obtain the graphic semantic features output by the initial graphic semantic determining layer. And synchronously updating parameters of the initial segmentation layer, the initial text semantic determination layer and the initial graphic semantic determination layer based on the output of the initial text semantic determination layer and the output of the initial graphic semantic determination layer to construct a loss function. And obtaining a trained segmentation layer, a text semantic determining layer and a graphic semantic determining layer through parameter updating.

In some embodiments, the text semantic feature and the graphic semantic feature may be respectively matched in the reference trademark data, a text semantic similarity risk coefficient of each reference trademark in the text semantic feature and the reference trademark data may be determined, and a graphic semantic similarity risk coefficient of each reference trademark in the reference trademark data may be determined; and carrying out weighting processing based on the text semantic similarity risk coefficient corresponding to the text semantic feature, the graphic semantic similarity risk coefficient corresponding to the graphic semantic feature and the sixth weight corresponding to each of the graphic semantic feature, and determining the semantic similarity risk coefficient.

The matching manner of determining the text semantic similarity risk coefficient is similar to that of determining the morphological similarity risk coefficient, and more description is given in fig. 2, and details are not repeated here.

The matching manner of determining the semantic similarity risk coefficient of the graph is similar to that of determining the morphological similarity risk coefficient, and more description is given in fig. 2, and details are not repeated here.

In some embodiments, the sixth weight corresponding to the text semantic feature may be positively correlated to the text saliency of the target trademark, and the sixth weight corresponding to the graphic semantic feature may be positively correlated to the graphic saliency of the target trademark.

Text saliency refers to the degree to which text in a target trademark plays a primary recognition role in the trademark.

The graphic saliency refers to the degree to which the graphics in the trademark play a major role in recognition in the trademark.

In some embodiments, text saliency and graphics saliency may be determined by a saliency determination model. Accordingly, the fourth label used for training the saliency determination model may further include a historical text saliency and a historical graphics saliency corresponding to the historical partial morphology features. See fig. 3 for more description of a saliency determination model.

In some embodiments of the present disclosure, the semantic similarity risk coefficient is obtained by weighting and calculating the graphic semantic feature based on the text semantic feature, the weight of the text semantic feature is related to the text saliency of the target trademark, and the weight of the graphic semantic feature is related to the graphic saliency of the target trademark, so that the determination of the semantic similarity risk coefficient is more accurate.

In some cases, some trademarks may not determine whether two trademarks belong to the same or similar trademarks only by morphology, and whether two trademarks belong to the same or similar trademarks may be determined by semantic features. For example, two trademarks respectively include only simplified or complex Chinese characters, have the same meaning, and have high probability of belonging to the same or similar trademarks. In some embodiments of the present description, the scope of the reference trademark (i.e., the same or similar trademark) may be further increased by semantics, thereby increasing the accuracy of the semantic similarity risk factor evaluation. And the semantic features of the target trademark are determined through the semantic feature determination model, and the semantic similarity risk coefficient is determined by matching the semantic features in the reference trademark data, so that the semantic similarity risk coefficient is determined more quickly and accurately.

Some embodiments of the present specification provide a brand risk screening device comprising at least one processor and at least one memory; the at least one memory is configured to store computer instructions; the at least one processor is configured to execute at least some of the computer instructions to implement the trademark risk screening method of any one of the above embodiments.

Some embodiments of the present disclosure provide a computer readable storage medium storing computer instructions that, when read by a computer, perform the trademark risk screening method of any one of the above embodiments.

It should be noted that the above description of the flow is only for the purpose of illustration and description, and does not limit the application scope of the present specification. Various modifications and changes to the flow may be made by those skilled in the art under the guidance of this specification. However, such modifications and variations are still within the scope of the present description.

While the basic concepts have been described above, it will be apparent to those skilled in the art that the foregoing detailed disclosure is by way of example only and is not intended to be limiting. Although not explicitly described herein, various modifications, improvements, and adaptations to the present disclosure may occur to one skilled in the art. Such modifications, improvements, and modifications are intended to be suggested within this specification, and therefore, such modifications, improvements, and modifications are intended to be included within the spirit and scope of the exemplary embodiments of the present invention.

Meanwhile, the specification uses specific words to describe the embodiments of the specification. Reference to "one embodiment," "an embodiment," and/or "some embodiments" means that a particular feature, structure, or characteristic is associated with at least one embodiment of the present description. Thus, it should be emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various positions in this specification are not necessarily referring to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of the present description may be combined as suitable.

Furthermore, the order in which the elements and sequences are processed, the use of numerical letters, or other designations in the description are not intended to limit the order in which the processes and methods of the description are performed unless explicitly recited in the claims. While certain presently useful inventive embodiments have been discussed in the foregoing disclosure, by way of various examples, it is to be understood that such details are merely illustrative and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements included within the spirit and scope of the embodiments of the present disclosure. For example, while the system components described above may be implemented by hardware devices, they may also be implemented solely by software solutions, such as installing the described system on an existing server or mobile device.

Likewise, it should be noted that in order to simplify the presentation disclosed in this specification and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof. This method of disclosure, however, is not intended to imply that more features than are presented in the claims are required for the present description. Indeed, less than all of the features of a single embodiment disclosed above.

In some embodiments, numbers describing the components, number of attributes are used, it being understood that such numbers being used in the description of embodiments are modified in some examples by the modifier "about," approximately, "or" substantially. Unless otherwise indicated, "about," "approximately," or "substantially" indicate that the number allows for a 20% variation. Accordingly, in some embodiments, numerical parameters set forth in the specification and claims are approximations that may vary depending upon the desired properties sought to be obtained by the individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and employ a method for preserving the general number of digits. Although the numerical ranges and parameters set forth herein are approximations that may be employed in some embodiments to confirm the breadth of the range, in particular embodiments, the setting of such numerical values is as precise as possible.

Each patent, patent application publication, and other material, such as articles, books, specifications, publications, documents, etc., referred to in this specification is incorporated herein by reference in its entirety. Except for application history documents that are inconsistent or conflicting with the content of this specification, documents that are currently or later attached to this specification in which the broadest scope of the claims to this specification is limited are also. It is noted that, if the description, definition, and/or use of a term in an attached material in this specification does not conform to or conflict with what is described in this specification, the description, definition, and/or use of the term in this specification controls.

Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments of this specification. Other variations are possible within the scope of this description. Thus, by way of example, and not limitation, alternative configurations of embodiments of the present specification may be considered as consistent with the teachings of the present specification. Accordingly, the embodiments of the present specification are not limited to only the embodiments explicitly described and depicted in the present specification.

Claims

1. A trademark risk investigation method based on big data, the method being implemented based on a trademark risk investigation system, the method comprising:

Acquiring a first data sample, registration classification and international classification of a target trademark;

determining at least one pre-estimated audit class based on the registration class and the international class;

acquiring the reference trademark data of the at least one estimated review category based on the big data;

determining target feature data of the target trademark through a machine learning model based on the first data sample;

matching in the reference trademark data based on the target characteristic data, and determining a similar risk coefficient of the target trademark;

and determining a violation risk coefficient of the target trademark through a judging database based on the target characteristic data, wherein the violation risk coefficient at least comprises a form violation risk coefficient.

2. The method of claim 1, wherein the similar risk factor comprises a morphological similar risk factor, the target feature data comprises a first morphological feature of the target trademark, the method comprising:

determining the first morphological feature through a morphological feature extraction model based on the first data sample, wherein the morphological feature extraction model is a machine learning model;

and based on the first morphological feature, matching in the reference trademark data, and determining the morphological similarity risk coefficient.

3. The method of claim 2, wherein the morphological similarity risk coefficient comprises an overall morphological similarity risk coefficient and a partial morphological similarity risk coefficient, and the morphological feature extraction model comprises a first feature extraction layer and a second feature extraction layer;

the determination mode of the overall form similarity risk coefficient comprises the following steps:

determining, based on the first data sample, overall morphological features of the target trademark through the first feature extraction layer;

based on the overall morphological characteristics, matching is carried out in the reference trademark data, and the overall morphological similarity risk coefficient is determined;

the determining mode of the at least one partial morphological similarity risk coefficient comprises the following steps:

determining at least one partial morphological feature of the target trademark by the second feature extraction layer based on the first data sample;

and determining the partial morphological similarity risk coefficient based on the matching of the at least one partial morphological feature in the reference trademark data.

4. The method of claim 1, wherein the similarity risk coefficient comprises a semantic similarity risk coefficient, the target feature data comprises a semantic feature of the target trademark, the method comprising:

Determining the semantic features through a semantic feature determining model based on the first data sample, wherein the semantic feature determining model is a machine learning model;

and determining the semantic similarity risk coefficient based on the semantic features matched in the reference trademark data.

5. The method according to claim 1, wherein the method further comprises: based on the offence risk factors and the similar risk factors, at least one preferred modification is recommended to the user.

6. A big data based trademark risk investigation system, the system comprising:

the first acquisition module is used for acquiring a first data sample, registration classification and international classification of the target trademark;

a first determining module for determining at least one pre-estimated review category based on the registration category and the international category;

the second acquisition module is used for acquiring the reference trademark data of the at least one estimated examination classification based on the big data;

a second determining module for

7. The system of claim 6, wherein the similarity risk factor comprises a morphological similarity risk factor, the target feature data comprises a first morphological feature of the target trademark, and the second determination module is further configured to:

8. The system of claim 6, wherein the similarity risk coefficient comprises a semantic similarity risk coefficient, the target feature data comprises a semantic feature of the target trademark, and the second determination module is further configured to:

9. A trademark risk screening device based on big data, characterized in that the device comprises at least one processor and at least one memory;

the at least one memory is configured to store computer instructions;

the at least one processor is configured to execute at least some of the computer instructions to implement the big data based brand risk screening method of any of claims 1 to 5.

10. A computer-readable storage medium storing computer instructions that, when read by a computer, perform the big data based trademark risk assessment method of any of claims 1-5.