Method and device for extracting features of advertisement copy

Abstract

The present invention discloses a method and device for extracting features from advertising copy, the method comprising: determining a plurality of preset text segments corresponding to a plurality of feature types; determining a target advertising copy; the target advertising copy includes a plurality of text segments; matching the plurality of text segments in the target advertising copy with the plurality of preset text segments to obtain a matching text segment in the target advertising copy corresponding to any of the feature types. It can be seen that the present invention can match the target advertising copy according to the preset text segments corresponding to a plurality of feature types to determine the text segments in the advertising copy corresponding to different feature types, thereby improving the accuracy of extracting feature information from the advertising copy, and providing an accurate data basis for the subsequent recognition or generation of feature information of the advertising copy.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method and device for extracting features from advertisement copy. Background Art

[0002] With the development of Internet technology, the forms of online advertisements have become more and more complex and diversified. In order to improve the efficiency of generating advertisement content, the existing technology has begun to introduce natural language processing technology and neural network technology into the field of algorithms for automatically generating advertisements. However, when determining the templates for generating advertisement copy, the existing technology generally only manually selects the iconic advertisement copy forms without considering the relevance of the advertisement copy itself. In addition, the manual method also has the problems of low efficiency and poor accuracy. It can be seen that the existing technology has defects that need to be solved urgently. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and device for extracting and determining features of advertising copy, which can improve the accuracy of extracting feature information of advertising copy and provide an accurate data basis for subsequent identification or generation of advertising copy feature information.

[0004] In order to solve the above technical problems, the first aspect of the present invention discloses a method for extracting features of advertising copy, the method comprising:

[0005] Determine a plurality of preset text segments corresponding to a plurality of feature types;

[0006] Determining a target advertising copy; the target advertising copy includes a plurality of text segments;

[0007] The multiple text segments in the target advertisement copy are matched with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types.

[0008] As an optional embodiment, in the first aspect of the present invention, the feature type includes at least one of the category of promotional object, the composition of promotional object, the efficacy of promotional object, the brand of promotional object and promotional marketing terms.

[0009] As an optional implementation manner, in the first aspect of the present invention, matching the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types includes:

[0010] Determining a preset text vector corresponding to each of the preset text segments;

[0011] Determine a matching text segment from all the text segments in the target advertisement copy, and determine a matching text vector of the matching text segment;

[0012] Determining a target preset text segment that matches the matching text segment among all the preset text segments according to similarity information between the matching text vector and the preset text vector;

[0013] The feature type corresponding to the matching text segment is determined as the feature type corresponding to the target preset text segment.

[0014] As an optional implementation manner, in the first aspect of the present invention, determining a matching text segment from all the text segments in the target advertisement copy includes:

[0015] Determine a part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule; the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments;

[0016] The text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy are determined as matching text segments.

[0017] As an optional implementation manner, in the first aspect of the present invention, the determining, according to the similarity information between the matching text vector and the preset text vector, a target preset text segment that matches the matching text segment among all the preset text segments comprises:

[0018] Based on a clustering algorithm, clustering calculation is performed on all the preset text vectors to obtain multiple preset text vector sets;

[0019] Determine, according to all the preset text vectors in each of the preset text vector sets, a set vector corresponding to each of the preset text vector sets;

[0020] Calculating similarity information between the matching text vector and a set vector corresponding to any one of the preset text vector sets, and determining whether the similarity information is greater than a first similarity threshold;

[0021] If the judgment result is yes, calculating the similarity information between the matching text vector and all the preset text vectors in the preset text vector set;

[0022] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector in the preset text vector set and / or higher than a second similarity threshold is determined as the target preset text segment.

[0023] As an optional embodiment, in the first aspect of the present invention, the similarity information includes at least one of cosine similarity information, Euclidean distance similarity information, Jaccard distance similarity information, edit distance similarity information, Chebyshev distance similarity information, Hamming distance similarity information, Mahalanobis distance similarity information, Manhattan distance similarity information and Minkowski distance similarity information.

[0024] As an optional implementation manner, in the first aspect of the present invention, before determining the target preset text segment that matches the matching text segment in all the preset text segments according to the similarity information between the matching text vector and the preset text vector, the method further includes:

[0025] Performing character matching between the target advertisement copy and the plurality of preset text segments to obtain a text segment in the target advertisement copy that is identical to any of the preset text segments;

[0026] Calculating the proportion information of all the same text segments in the target advertisement copy, and determining whether the proportion information is lower than a proportion threshold;

[0027] If the judgment result is yes, the step of determining a target preset text segment that matches the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector is performed;

[0028] And, the method further comprises:

[0029] The same text segment is determined as the matching text segment, and the feature type of the same text segment is determined as the feature type corresponding to the corresponding preset text segment.

[0030] As an optional embodiment, in the first aspect of the present invention, the method further comprises:

[0031] Determine the blacklist text data corresponding to each of the feature types;

[0032] For all matching text segments corresponding to any of the feature types in the target advertising copy, a feature association cancellation operation is performed on the matching text segments in the blacklist text data corresponding to the feature type; the feature association cancellation operation is used to cancel the association between the matching text segment and its corresponding feature type.

[0033] The second aspect of the present invention discloses a feature extraction device for advertising copy, the device comprising:

[0034] A first determination module is used to determine a plurality of preset text segments corresponding to a plurality of feature types;

[0035] A second determination module is used to determine a target advertisement copy; the target advertisement copy includes a plurality of text segments;

[0036] A matching module is used to match the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types.

[0037] As an optional implementation, in the second aspect of the present invention, the feature type includes at least one of the category of promotional object, the composition of promotional object, the efficacy of promotional object, the brand of promotional object and promotional marketing terms.

[0038] As an optional implementation, in the second aspect of the present invention, the matching module includes:

[0039] A first determining unit, configured to determine a preset text vector corresponding to each of the preset text segments;

[0040] A second determining unit, configured to determine a matching text segment from all the text segments in the target advertisement copy, and determine a matching text vector of the matching text segment;

[0041] A third determining unit, configured to determine a target preset text segment that matches the matching text segment among all the preset text segments according to similarity information between the matching text vector and the preset text vector;

[0042] The fourth determining unit is used to determine the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment.

[0043] As an optional implementation, in the second aspect of the present invention, the second determination unit determines a specific manner of matching text segments from all the text segments in the target advertisement copy, including:

[0044] Determine a part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule; the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments;

[0045] The text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy are determined as matching text segments.

[0046] As an optional implementation, in the second aspect of the present invention, the third determination unit determines the specific manner of the target preset text segment matching the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector, including:

[0047] Based on a clustering algorithm, clustering calculation is performed on all the preset text vectors to obtain multiple preset text vector sets;

[0048] Determine, according to all the preset text vectors in each of the preset text vector sets, a set vector corresponding to each of the preset text vector sets;

[0049] Calculating similarity information between the matching text vector and a set vector corresponding to any one of the preset text vector sets, and determining whether the similarity information is greater than a first similarity threshold;

[0050] If the judgment result is yes, calculating the similarity information between the matching text vector and all the preset text vectors in the preset text vector set;

[0051] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector in the preset text vector set and / or higher than a second similarity threshold is determined as the target preset text segment.

[0052] As an optional embodiment, in the second aspect of the present invention, the similarity information includes at least one of cosine similarity information, Euclidean distance similarity information, Jaccard distance similarity information, edit distance similarity information, Chebyshev distance similarity information, Hamming distance similarity information, Mahalanobis distance similarity information, Manhattan distance similarity information and Minkowski distance similarity information.

[0053] As an optional implementation, in the second aspect of the present invention, the device further includes a proportion judgment module, which is used to perform the following steps before the third determination unit determines the target preset text segment that matches the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector:

[0054] Performing character matching between the target advertisement copy and the plurality of preset text segments to obtain a text segment in the target advertisement copy that is identical to any of the preset text segments;

[0055] Calculating the proportion information of all the same text segments in the target advertisement copy, and determining whether the proportion information is lower than a proportion threshold;

[0056] If the judgment result is yes, the third determination unit is triggered to execute the step of determining a target preset text segment that matches the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector;

[0057] And, the device further comprises:

[0058] The third determination module is used to determine the same text segment as the matching text segment, and determine the feature type of the same text segment as the feature type corresponding to the corresponding preset text segment.

[0059] As an optional implementation, in the second aspect of the present invention, the device further includes a blacklist determination module, configured to perform the following steps:

[0060] Determine the blacklist text data corresponding to each of the feature types;

[0061] For all matching text segments corresponding to any of the feature types in the target advertising copy, a feature association cancellation operation is performed on the matching text segments in the blacklist text data corresponding to the feature type; the feature association cancellation operation is used to cancel the association between the matching text segment and its corresponding feature type.

[0062] The third aspect of the present invention discloses another device for extracting features from advertising texts, the device comprising:

[0063] A memory storing executable program code;

[0064] a processor coupled to the memory;

[0065] The processor calls the executable program code stored in the memory to execute part or all of the steps in the method for extracting features of advertising text disclosed in the first aspect of the embodiment of the present invention.

[0066] A fourth aspect of an embodiment of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all of the steps in the feature extraction method of advertising copy disclosed in the first aspect of the embodiment of the present invention.

[0067] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0068] In an embodiment of the present invention, a plurality of preset text segments corresponding to a plurality of feature types are determined; a target advertisement copy is determined; the target advertisement copy includes a plurality of text segments; the plurality of text segments in the target advertisement copy are matched with the plurality of preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types. It can be seen that the present invention can match the target advertisement copy according to the preset text segments corresponding to a plurality of feature types to determine the text segments in the advertisement copy corresponding to different feature types, thereby improving the accuracy of extracting feature information of the advertisement copy, and can provide an accurate data basis for the subsequent recognition or generation of feature information of the advertisement copy. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0070] Figure 1 It is a flow chart of a method for extracting features of advertisement text disclosed in an embodiment of the present invention;

[0071] Figure 2 It is a flow chart of another method for extracting features of advertisement text disclosed in an embodiment of the present invention;

[0072] Figure 3 It is a structural schematic diagram of a feature extraction device for advertisement text disclosed in an embodiment of the present invention;

[0073] Figure 4 It is a structural schematic diagram of another device for extracting features of advertisement text disclosed in an embodiment of the present invention;

[0074] Figure 5 It is a structural schematic diagram of another advertising copy feature extraction device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0077] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0078] The present invention discloses a method and device for extracting features of advertisement copy, which can match target advertisement copy according to preset text segments corresponding to multiple feature types, so as to determine text segments corresponding to different feature types in the advertisement copy, thereby improving the accuracy of extracting feature information of the advertisement copy, and providing an accurate data basis for the subsequent recognition or generation of feature information of the advertisement copy. The following is a detailed description of each.

[0079] Embodiment 1

[0080] See also Figure 1 , Figure 1 1 is a flow chart of a method for extracting features of an advertisement text disclosed in an embodiment of the present invention. Figure 1 The described method is applied to a document text processing device, which may be a corresponding processing terminal, processing equipment or processing server, and the server may be a local server or a cloud server, which is not limited in the embodiments of the present invention. Figure 1 As shown, the feature extraction method of the advertisement copy may include the following operations:

[0081] 101. Determine multiple preset text segments corresponding to multiple feature types.

[0082] In an embodiment of the present invention, the feature type may include at least one of the category of the promotional object, the composition of the promotional object, the efficacy of the promotional object, the brand of the promotional object, and the promotional marketing terms. Optionally, the method of determining the multiple preset text segments corresponding to the multiple feature types may be to manually screen the text segments of the multiple feature types in the existing advertising copy to obtain the multiple preset text segments corresponding to the multiple feature types, or to automatically identify the multiple preset text segments corresponding to the multiple feature types in the preset advertising copy through a neural network algorithm. Optionally, multiple preset text segments corresponding to the multiple feature types may be screened out from the promotional copy of each network platform. For example, multiple preset text segments such as "mobile phone screen film", "high performance", "foldable design", "anti-fingerprint", "anti-scratch", "large coupons" and "limited time flash sale" may be obtained from the promotional copy of each first-class e-commerce and second-class e-commerce platform, and the feature types corresponding to different preset text segments may be identified, such as "mobile phone screen film" belongs to the type of the promotional object category, while "high performance", "foldable design", "anti-fingerprint" or "anti-scratch" are all types of the promotional object efficacy, and "large coupons" and "limited time flash sale" are types of promotional marketing terms.

[0083] 102. Determine the target advertising copy.

[0084] Optionally, the target advertisement copy includes multiple text fragments. Optionally, the advertisement copy may be determined by crawling the text of a specific network address, or by receiving advertisement copy input after manual screening, or by obtaining advertisement text by means of voice recognition or image recognition, which is not limited by the present invention. Optionally, the advertisement copy may be promotional copy for goods or services of different categories, such as promotional copy used in the introduction page of goods or services or in business activities, or may be promotional copy used to attract traffic to goods or services, such as article titles or story quotes written by specific media such as WeChat public accounts to attract users, which is not limited by the present invention. Optionally, the language of the advertisement copy is not limited to Chinese, and may also be other grammatically clear languages ​​such as English and French, which is not limited by the present invention.

[0085] 103. Match multiple text segments in the target advertisement copy with multiple preset text segments to obtain a matching text segment corresponding to any feature type in the target advertisement copy.

[0086] Optionally, multiple text segments in the target advertising copy are matched with multiple preset text segments, which can be a one-to-one character match. In this case, when a text segment in the target advertising copy is matched to have characters that are completely or partially identical to those of the preset text segment, for example, when the number of identical characters is higher than a preset character count threshold, the text segment can be determined as a matching text segment, and the corresponding feature type can be determined as the feature type corresponding to the preset text segment. Optionally, multiple text segments in the target advertising copy are matched with multiple preset text segments, which can be a word vector match. In this case, it is necessary to calculate the similarity between the vector corresponding to the text segment in the target advertising copy and the vector corresponding to the preset text segment to determine whether the two match. In the case of a match, the text segment can be determined as a matching text segment, and the corresponding feature type can be determined as the feature type corresponding to the preset text segment.

[0087] Optionally, the matching text segments corresponding to any feature type in the target advertising copy are determined and subsequently used to train a neural network model to recognize the feature information of the target advertising copy. They can also be used in an automatic copy generation algorithm as reference data or learning data for generating copy feature information, which is not limited by the present invention.

[0088] It can be seen that the method described in the embodiment of the present invention can match the target advertising copy according to the preset text segments corresponding to multiple feature types to determine the text segments in the advertising copy corresponding to different feature types, thereby improving the accuracy of extracting the feature information of the advertising copy, and providing an accurate data basis for the subsequent identification or generation of the advertising copy feature information.

[0089] As an optional implementation, in the above step 103, matching multiple text segments in the target advertisement copy with multiple preset text segments to obtain a matching text segment corresponding to any feature type in the target advertisement copy includes:

[0090] Determine a preset text vector corresponding to each preset text segment;

[0091] Determine a matching text segment from all text segments in the target advertisement copy, and determine a matching text vector of the matching text segment;

[0092] Determine, according to the similarity information between the matching text vector and the preset text vector, a target preset text segment that matches the matching text segment among all preset text segments;

[0093] The feature type corresponding to the matching text segment is determined as the feature type corresponding to the target preset text segment.

[0094] In the embodiment of the present invention, the method of determining the preset text vector or determining the matching text vector can be to convert the text to the vector using a pre-trained word vector model, for example, the open source Tencent word vector model and Sogou word vector model can be used, or a fixed representation model based on word vectors such as fasttext, word2vec, glove, etc. can be used. Optionally, the matching text vector of the matching text segment can be determined by calculating the superposition of the word vectors corresponding to all the divisible text segments in each matching text segment to determine the matching text vector. For example, the full-mode word segmentation function of the Jieba word segmentation tool can be used to segment the matching text segment to obtain multiple sub-segments, and then the word vector corresponding to each sub-segment can be found in the Tencent word vector list, and then the word vectors of each sub-segment are superimposed as the representation vector of the text segment, that is, the matching text vector. An example of full-mode word segmentation is to perform full-mode word segmentation on "I came to Tsinghua University in Beijing" to obtain multiple sub-segments such as I / come / Beijing / Tsinghua / Tsinghua University / Hua Da / University.

[0095] Optionally, determining a target preset text segment that matches the matching text segment among all preset text segments according to similarity information between the matching text vector and the preset text vector may include:

[0096] For a matching text vector corresponding to any matching text segment, calculating similarity information between the matching text vector and any preset text vector;

[0097] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector and / or having similarity information greater than a third similarity threshold among all preset text vectors is determined as the target preset text segment.

[0098] Optionally, the expression of the highest similarity information is used to define the highest similarity between the two vectors, and the expression of the similarity information being greater than the third similarity threshold is used to define the similarity between the two vectors being greater than a preset degree. Both are not limited to the comparison of numerical values. This is because in some cases, the higher the value of the similarity information, the lower the similarity. At this time, the meaning of the highest similarity information is that the value of the similarity information is the lowest, and the meaning of the similarity information being greater than the third similarity threshold is that the value of the similarity information is less than the third similarity threshold. In most cases, the higher the value of the similarity information, the higher the similarity. At this time, the meaning of the highest similarity information is that the value of the similarity information is the highest, and the meaning of the similarity information being greater than the third similarity threshold is that the value of the similarity information is greater than the third similarity threshold. Optionally, in this embodiment, the setting of the third similarity threshold can be set according to experimental values ​​or empirical values.

[0099] Optionally, the similarity information may include at least one of cosine similarity information, Euclidean distance similarity information, Jaccard distance similarity information, edit distance similarity information, Chebyshev distance similarity information, Hamming distance similarity information, Mahalanobis distance similarity information, Manhattan distance similarity information, and Minkowski distance similarity information. Optionally, when the similarity information includes a combination of at least two of the above multiple similarity information types, the similarity information may be a weighted combination of multiple similarity information types, wherein different similarity information types may have different weights. In a preferred embodiment, the similarity information is cosine similarity information.

[0100] It can be seen that by implementing this optional implementation, it is possible to determine the target preset text segment that matches the matching text segment among all preset text segments based on the similarity information between the matching text vector and the preset text vector, and determine the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment, thereby accurately extracting text segments corresponding to different feature types in the advertising copy, and providing an accurate data basis for the subsequent identification or generation of advertising copy feature information.

[0101] As an optional implementation manner, in the above step, determining the matching text segment from all text segments in the target advertisement copy includes:

[0102] Determine part-of-speech combination rules based on multiple preset text fragments and part-of-speech tagging rules;

[0103] The text fragments that meet the part-of-speech combination rules among all the text fragments in the target advertisement copy are determined as matching text fragments.

[0104] Optionally, the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments. Optionally, determining the part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule may include:

[0105] According to the part-of-speech tagging algorithm, each preset text segment is tagged with the part-of-speech to obtain multiple part-of-speech combinations;

[0106] The number of occurrences of each part-of-speech combination is calculated, and the part-of-speech combinations whose number of occurrences is higher than a preset number threshold are retained to obtain a set of part-of-speech combinations.

[0107] Optionally, determining the text segments that meet the part-of-speech combination rules among all the text segments in the target advertisement copy as matching text segments may include:

[0108] According to the part-of-speech tagging algorithm, each text segment in the target advertisement copy is tagged with a part-of-speech to obtain a plurality of corresponding text part-of-speech combinations;

[0109] Any text part-of-speech combination is matched with all part-of-speech combinations in a part-of-speech combination set. When the matching result shows that there is a part-of-speech combination in the part-of-speech combination set that is identical to the text part-of-speech combination, the text segment corresponding to the text part-of-speech combination is determined as the matching text segment.

[0110] Optionally, the part-of-speech tagging algorithm can be a rule-based part-of-speech tagging algorithm, a statistical model-based part-of-speech tagging algorithm, a part-of-speech tagging algorithm based on a combination of statistical methods and rule methods, or a part-of-speech tagging algorithm based on deep learning, such as Jieba part-of-speech tagging tool, SnowNLP part-of-speech tagging tool, THULAC part-of-speech tagging tool, StanfordCoreNLP part-of-speech tagging tool, HanLP part-of-speech tagging tool, NLTK part-of-speech tagging platform, or SpaCy part-of-speech tagging tool.

[0111] For example, in a specific implementation scheme, the part-of-speech tagging function in the Jieba word segmentation tool can be used to annotate the part-of-speech of each phrase in a fragment list including multiple preset text fragments corresponding to multiple feature types. The preset text fragments in the fragment list are converted into part-of-speech combinations, and a threshold is set for the number of occurrences. Part-of-speech combinations above the threshold are extracted. For example, "clear and delicate picture quality" corresponds to n(2)+adj(2)+adj(2), where the number of characters included is in brackets. At the same time, each part-of-speech combination is classified according to the feature type corresponding to the corresponding preset text fragment. Subsequently, when determining the matching text fragment, the part-of-speech combination can be used to match the target advertising copy, and the text fragments in the target advertising copy that meet the part-of-speech combination rules can be extracted.

[0112] It can be seen that by implementing this optional implementation method, it is possible to determine the part-of-speech combination rules based on multiple preset text fragments and part-of-speech tagging rules, and determine the text fragments that meet the part-of-speech combination rules among all text fragments in the target advertising copy as matching text fragments, so that the text fragments to be subsequently vector matched can be accurately determined through the part-of-speech rules of the preset text, thereby facilitating more accurate extraction of feature text fragments in the advertising copy and providing an accurate data basis for the subsequent identification or generation of feature information of the advertising copy.

[0113] As an optional implementation, in the above step, determining the target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector may include:

[0114] Based on a clustering algorithm, clustering calculation is performed on all preset text vectors to obtain multiple preset text vector sets;

[0115] Determine, according to all preset text vectors in each preset text vector set, a set vector corresponding to each preset text vector set;

[0116] Calculate similarity information between the matching text vector and a set vector corresponding to any preset text vector set, and determine whether the similarity information is greater than a first similarity threshold;

[0117] If the judgment result is yes, calculate the similarity information between the matching text vector and all the preset text vectors in the preset text vector set;

[0118] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector in the preset text vector set and / or higher than the second similarity threshold is determined as the target preset text segment.

[0119] Optionally, the clustering algorithm may be hierarchical clustering, k-means algorithm, EM algorithm, DBSCAN algorithm, OPTICS algorithm, Mean Shift algorithm or spectral clustering algorithm. Optionally, the meaning of the related expressions of the similarity information and the first similarity threshold and the second similarity threshold in this embodiment and the related specific technical details may refer to the expressions of the similarity threshold and the third similarity threshold in the above embodiment, which will not be repeated here.

[0120] Optionally, based on all the preset text vectors in each preset text vector set, the set vector corresponding to each preset text vector set can be determined by taking the sum and averaging all the preset text vectors in each preset text vector set to determine the set vector corresponding to the preset text vector set. For example, in a specific implementation scheme, all preset text fragments can be converted into a word vector list using the Tencent word vector model, and then each word vector in the word vector list can be clustered using the k-means clustering algorithm to divide the word vectors into different clusters, and the cluster representation vector of each cluster is obtained by summing and averaging the word vectors of all the words in the cluster. Each cluster is a word vector set.

[0121] It can be seen that by implementing this optional implementation, the similarity information between the matching text vector and the set vector corresponding to any preset text vector set can be calculated first, and then the similarity information between the matching text vector and all preset text vectors in the sufficiently similar preset text vector set can be calculated to determine the target preset text segment, thereby effectively reducing the amount of calculation, making the vector matching operation more efficient and simple, and achieving good algorithm efficiency when matching vectors with large amounts of data.

[0122] As an optional implementation, in the above step, before determining the target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector, the method further includes:

[0123] Matching characters in the target advertisement copy with multiple preset text segments to obtain a text segment in the target advertisement copy that is identical to any of the preset text segments;

[0124] Calculate the proportion of all identical text segments in the target advertisement copy, and determine whether the proportion is lower than a proportion threshold;

[0125] If the judgment result is yes, the above step of determining a target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector is executed.

[0126] Optionally, the proportion of all identical text segments in the target advertising copy can be the ratio of the sum of the number of characters of all identical text segments to the total number of characters of the target advertising copy. In a specific implementation scheme, the number of words D of the target advertising copy and the number of words E of all identical text segments can be counted, and the value F of E / D can be calculated. If F is too small, it means that the identified information point is too small, and the length of the advertising copy is too large. At this time, there is a high probability that some text segments are not completely the same as the preset text segments. At this time, it is necessary to identify the feature text that has not appeared. In order to achieve the purpose of this judgment, a threshold value G related to F can be introduced. If F is less than G, the next step of vector matching is entered to determine the target preset text segment that matches the matching text segment among all preset text segments based on the similarity information between the matching text vector and the preset text vector.

[0127] It can be seen that by implementing this optional implementation method, it is possible to calculate the proportion information of all identical text fragments in the target advertising copy, and determine whether the proportion information is lower than the proportion threshold, so as to determine whether the feature text obtained by character matching is sufficient to represent the feature information in the advertising copy, and perform a vector matching operation when it is determined that it cannot be represented, so as to facilitate the subsequent more accurate extraction of the feature text fragments in the advertising copy, and provide an accurate data basis for the subsequent identification or generation of advertising copy feature information.

[0128] As an optional implementation, the method further includes:

[0129] The same text segment is determined as a matching text segment, and the feature type of the same text segment is determined as a feature type corresponding to the corresponding preset text segment.

[0130] It can be seen that by implementing this optional implementation, the same text segment matched by characters can also be confirmed as the characteristic text in the copy, providing an accurate data basis for the subsequent recognition or generation of characteristic information of the advertising copy.

[0131] Embodiment 2

[0132] See also Figure 2 , Figure 2 1 is a flow chart of another method for extracting features of advertisement text disclosed in an embodiment of the present invention. Figure 2 The described method is applied to a document text processing device, which may be a corresponding processing terminal, processing equipment or processing server, and the server may be a local server or a cloud server, which is not limited in the embodiments of the present invention. Figure 2 As shown, the feature extraction method of the advertisement copy may include the following operations:

[0133] 201. Determine multiple preset text segments corresponding to multiple feature types.

[0134] 202. Determine the target advertising copy.

[0135] 203. Match multiple text segments in the target advertisement copy with multiple preset text segments to obtain a matching text segment corresponding to any feature type in the target advertisement copy.

[0136] The specific technical details and technical terminology of the above steps 201-203 can be referred to the description of steps 101-103 in the first implementation, and will not be repeated here.

[0137] 204. Determine the blacklist text data corresponding to each feature type.

[0138] Optionally, the blacklist text data corresponding to each feature type can be determined by manually extracting the preset text fragments that cannot accurately characterize the feature type from all preset text fragments corresponding to a certain feature type obtained by screening the promotional copy of the network platform, and determining them as the blacklist text data corresponding to the feature type. Among them, the preset text fragment that cannot accurately characterize the feature type, that is, the text fragment in the blacklist text data corresponding to the feature type, can be a text fragment that has a high similarity with other text fragments that can accurately characterize the feature type but is easy to cause ambiguity. For example, in a specific embodiment, the blacklist text data of the promotional object category [BB cream / CC cream / air cushion] can be determined as [suit / makeup brush / shoes / comb / nude makeup cream / bed / eyebrow powder], so as to facilitate the subsequent screening and investigation of matching text fragments.

[0139] 205 . For all matching text segments corresponding to any feature type in the target advertisement copy, a feature association cancellation operation is performed for the matching text segments in the blacklist text data corresponding to the feature type.

[0140] Specifically, the feature association cancellation operation is used to cancel the association between the matching text fragment and its corresponding feature type. Specifically, the feature association cancellation operation can also be used to delete the matching text fragment in the blacklist text data corresponding to the feature type or cancel its identification as a matching text fragment. Optionally, the blacklist text data can also correspond to multiple feature types at the same time. For example, the blacklist text data can correspond to two feature types. When matching, the matching text fragment corresponding to the two feature types will be used to match the blacklist text data corresponding to the two feature types at the same time. For example, in a specific implementation scheme, a blacklist for the promotion object category can be constructed according to all preset text fragments, and then a blacklist of different promotion object ingredients, promotion object efficacy or promotion object brands under different promotion object categories can be constructed, and the target advertising copy can be matched according to these blacklists. The matching order is to first identify the promotion object category and then call the blacklist of the promotion object ingredient, promotion object efficacy or promotion object brand of the corresponding promotion object category to filter out the matching text fragment in the blacklist, so as to avoid the subsequent misinterpretation of the information related to the feature type of the copy when the copy is generated or the copy feature information is identified.

[0141] It can be seen that the method described in the embodiment of the present invention can cancel the feature association operation on the matching text fragments in the blacklist text data in the target advertising copy, thereby canceling the feature association of the matching text fragments that cannot accurately represent the corresponding feature type, so as to avoid subsequent distortion of the information related to the feature type of the copy, which is conducive to improving the accuracy of feature information extraction of the advertising copy.

[0142] Embodiment 3

[0143] See also Figure 3 , Figure 3 Schematic diagram of a feature extraction device for advertisement text disclosed in an embodiment of the present invention. Figure 3 The described device can be applied to a corresponding document text processing device, which can be a corresponding processing terminal, processing equipment or processing server, and the server can be a local server or a cloud server, which is not limited in the embodiment of the present invention. Figure 3 As shown, the device may include:

[0144] The first determining module 301 is used to determine a plurality of preset text segments corresponding to a plurality of feature types.

[0145] In an embodiment of the present invention, the feature type may include at least one of the category of the promotional object, the composition of the promotional object, the efficacy of the promotional object, the brand of the promotional object, and the promotional marketing terms. Optionally, the method of determining the multiple preset text segments corresponding to the multiple feature types may be to manually screen the text segments of the multiple feature types in the existing advertising copy to obtain the multiple preset text segments corresponding to the multiple feature types, or to automatically identify the multiple preset text segments corresponding to the multiple feature types in the preset advertising copy through a neural network algorithm.

[0146] The second determination module 302 is used to determine the target advertisement copy.

[0147] Optionally, the target advertisement copy includes multiple text fragments. Optionally, the advertisement copy may be determined by crawling the text of a specific network address, or by receiving advertisement copy input after manual screening, or by obtaining advertisement text by means of voice recognition or image recognition, which is not limited by the present invention. Optionally, the advertisement copy may be promotional copy for goods or services of different categories, such as promotional copy used in the introduction page of goods or services or in business activities, or may be promotional copy used to attract traffic to goods or services, such as article titles or story quotes written by specific media such as WeChat public accounts to attract users, which is not limited by the present invention. Optionally, the language of the advertisement copy is not limited to Chinese, and may also be other grammatically clear languages ​​such as English and French, which is not limited by the present invention.

[0148] The matching module 303 is used to match the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment corresponding to any feature type in the target advertisement copy.

[0149] Optionally, multiple text segments in the target advertising copy are matched with multiple preset text segments, which can be a one-to-one character match. In this case, when a text segment in the target advertising copy is matched to have characters that are completely or partially identical to those of the preset text segment, for example, when the number of identical characters is higher than a preset character count threshold, the text segment can be determined as a matching text segment, and the corresponding feature type can be determined as the feature type corresponding to the preset text segment. Optionally, multiple text segments in the target advertising copy are matched with multiple preset text segments, which can be a word vector match. In this case, it is necessary to calculate the similarity between the vector corresponding to the text segment in the target advertising copy and the vector corresponding to the preset text segment to determine whether the two match. In the case of a match, the text segment can be determined as a matching text segment, and the corresponding feature type can be determined as the feature type corresponding to the preset text segment.

[0150] Optionally, the matching text segments corresponding to any feature type in the target advertising copy are determined and subsequently used to train a neural network model to recognize the feature information of the target advertising copy. They can also be used in an automatic copy generation algorithm as reference data or learning data for generating copy feature information, which is not limited by the present invention.

[0151] It can be seen that the device described in the embodiment of the present invention can match the target advertising copy according to the preset text segments corresponding to multiple feature types to determine the text segments in the advertising copy corresponding to different feature types, thereby improving the accuracy of extracting the feature information of the advertising copy and providing an accurate data basis for the subsequent identification or generation of the feature information of the advertising copy.

[0152] As an optional implementation, Figure 4 As shown, the matching module 303 includes:

[0153] A first determining unit 3031 is used to determine a preset text vector corresponding to each preset text segment;

[0154] The second determining unit 3032 is used to determine a matching text segment from all text segments in the target advertisement copy, and determine a matching text vector of the matching text segment;

[0155] The third determining unit 3033 is used to determine a target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector;

[0156] The fourth determining unit 3034 is used to determine the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment.

[0157] In the embodiment of the present invention, the method of determining the preset text vector or determining the matching text vector can be to convert the text to the vector using a pre-trained word vector model, for example, the open source Tencent word vector model and Sogou word vector model can be used, or a fixed representation model based on word vectors such as fasttext, word2vec, glove, etc. can be used. Optionally, the matching text vector of the matching text segment can be determined by calculating the superposition of the word vectors corresponding to all the divisible text segments in each matching text segment to determine the matching text vector. For example, the full-mode word segmentation function of the Jieba word segmentation tool can be used to segment the matching text segment to obtain multiple sub-segments, and then the word vector corresponding to each sub-segment can be found in the Tencent word vector list, and then the word vectors of each sub-segment are superimposed as the representation vector of the text segment, that is, the matching text vector. An example of full-mode word segmentation is to perform full-mode word segmentation on "I came to Tsinghua University in Beijing" to obtain multiple sub-segments such as I / come / Beijing / Tsinghua / Tsinghua University / Hua Da / University.

[0158] Optionally, the specific manner in which the third determining unit 3033 determines the target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector may include:

[0159] For a matching text vector corresponding to any matching text segment, calculating similarity information between the matching text vector and any preset text vector;

[0160] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector and / or having similarity information greater than a third similarity threshold among all preset text vectors is determined as the target preset text segment.

[0161] Optionally, the expression of the highest similarity information is used to define the highest similarity between the two vectors, and the expression of the similarity information being greater than the third similarity threshold is intended to define the similarity between the two vectors being greater than a preset degree. Both are not limited to the comparison of numerical values. This is because in some cases, the higher the value of the similarity information, the lower the similarity. At this time, the highest similarity information means that the value of the similarity information is the lowest, and the similarity information being greater than the third similarity threshold means that the value of the similarity information is less than the third similarity threshold. In most cases, the higher the value of the similarity information, the higher the similarity. At this time, the highest similarity information means that the value of the similarity information is the highest, and the similarity information being greater than the third similarity threshold means that the value of the similarity information is greater than the third similarity threshold. Optionally, in this embodiment, the setting of the third similarity threshold can be set according to experimental values ​​or empirical values.

[0162] Optionally, the similarity information may include at least one of cosine similarity information, Euclidean distance similarity information, Jaccard distance similarity information, edit distance similarity information, Chebyshev distance similarity information, Hamming distance similarity information, Mahalanobis distance similarity information, Manhattan distance similarity information, and Minkowski distance similarity information. Optionally, when the similarity information includes a combination of at least two of the above multiple similarity information types, the similarity information may be a weighted combination of multiple similarity information types, wherein different similarity information types may have different weights. In a preferred embodiment, the similarity information is cosine similarity information.

[0163] It can be seen that by implementing this optional implementation, it is possible to determine the target preset text segment that matches the matching text segment among all preset text segments based on the similarity information between the matching text vector and the preset text vector, and determine the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment, thereby accurately extracting text segments corresponding to different feature types in the advertising copy, and providing an accurate data basis for the subsequent identification or generation of advertising copy feature information.

[0164] As an optional implementation manner, the second determining unit 3032 determines a specific way of matching text segments from all text segments in the target advertisement copy, including:

[0165] Determine a part-of-speech combination rule according to a plurality of preset text segments and a part-of-speech tagging rule; the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments;

[0166] The text fragments that meet the part-of-speech combination rules among all the text fragments in the target advertisement copy are determined as matching text fragments.

[0167] Optionally, the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments. Optionally, determining the part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule may include:

[0168] According to the part-of-speech tagging algorithm, each preset text segment is tagged with the part-of-speech to obtain multiple part-of-speech combinations;

[0169] The number of occurrences of each part-of-speech combination is calculated, and the part-of-speech combinations whose number of occurrences is higher than a preset number threshold are retained to obtain a set of part-of-speech combinations.

[0170] Optionally, the specific manner in which the second determining unit 3032 determines the text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy as matching text segments may include:

[0171] According to the part-of-speech tagging algorithm, each text segment in the target advertisement copy is tagged with a part-of-speech to obtain a plurality of corresponding text part-of-speech combinations;

[0172] Any text part-of-speech combination is matched with all part-of-speech combinations in a part-of-speech combination set. When the matching result shows that there is a part-of-speech combination in the part-of-speech combination set that is identical to the text part-of-speech combination, the text segment corresponding to the text part-of-speech combination is determined as the matching text segment.

[0173] Optionally, the part-of-speech tagging algorithm can be a rule-based part-of-speech tagging algorithm, a statistical model-based part-of-speech tagging algorithm, a part-of-speech tagging algorithm based on a combination of statistical methods and rule methods, or a part-of-speech tagging algorithm based on deep learning, such as Jieba part-of-speech tagging tool, SnowNLP part-of-speech tagging tool, THULAC part-of-speech tagging tool, StanfordCoreNLP part-of-speech tagging tool, HanLP part-of-speech tagging tool, NLTK part-of-speech tagging platform, or SpaCy part-of-speech tagging tool.

[0174] It can be seen that by implementing this optional implementation method, it is possible to determine the part-of-speech combination rules based on multiple preset text fragments and part-of-speech tagging rules, and determine the text fragments that meet the part-of-speech combination rules among all text fragments in the target advertising copy as matching text fragments, so that the text fragments to be subsequently vector matched can be accurately determined through the part-of-speech rules of the preset text, thereby facilitating more accurate extraction of feature text fragments in the advertising copy and providing an accurate data basis for the subsequent identification or generation of feature information of the advertising copy.

[0175] As an optional implementation manner, the third determining unit 3033 determines the specific manner of the target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector, including:

[0176] Based on a clustering algorithm, clustering calculation is performed on all preset text vectors to obtain multiple preset text vector sets;

[0177] Determine, according to all preset text vectors in each preset text vector set, a set vector corresponding to each preset text vector set;

[0178] Calculate similarity information between the matching text vector and a set vector corresponding to any preset text vector set, and determine whether the similarity information is greater than a first similarity threshold;

[0179] If the judgment result is yes, calculate the similarity information between the matching text vector and all the preset text vectors in the preset text vector set;

[0180] The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector in the preset text vector set and / or higher than the second similarity threshold is determined as the target preset text segment.

[0181] Optionally, the clustering algorithm may be hierarchical clustering, k-means algorithm, EM algorithm, DBSCAN algorithm, OPTICS algorithm, Mean Shift algorithm or spectral clustering algorithm. Optionally, the meaning of the related expressions of the similarity information and the first similarity threshold and the second similarity threshold in this embodiment and the related specific technical details may refer to the expressions of the similarity threshold and the third similarity threshold in the above embodiment, which will not be repeated here.

[0182] Optionally, based on all preset text vectors in each preset text vector set, the set vector corresponding to each preset text vector set is determined. The set vector corresponding to the preset text vector set can be determined by taking the average of the sum of all preset text vectors in each preset text vector set.

[0183] It can be seen that by implementing this optional implementation, the similarity information between the matching text vector and the set vector corresponding to any preset text vector set can be calculated first, and then the similarity information between the matching text vector and all preset text vectors in the sufficiently similar preset text vector set can be calculated to determine the target preset text segment, thereby effectively reducing the amount of calculation, making the vector matching operation more efficient and simple, and achieving good algorithm efficiency when matching vectors with large amounts of data.

[0184] As an optional implementation, Figure 4 As shown, the device also includes a proportion judgment module 304, which is used to perform the following steps before the third determination unit 3033 determines the target preset text segment that matches the matching text segment in all preset text segments according to the similarity information between the matching text vector and the preset text vector:

[0185] Matching characters in the target advertisement copy with multiple preset text segments to obtain a text segment in the target advertisement copy that is identical to any of the preset text segments;

[0186] Calculate the proportion of all identical text segments in the target advertisement copy, and determine whether the proportion is lower than a proportion threshold;

[0187] If the judgment result is yes, the third determining unit 3033 is triggered to execute the step of determining a target preset text segment that matches the matching text segment among all preset text segments according to the similarity information between the matching text vector and the preset text vector;

[0188] Optionally, the proportion of all identical text segments in the target advertising copy can be the ratio of the sum of the number of characters of all identical text segments to the total number of characters of the target advertising copy. In a specific implementation scheme, the number of words D of the target advertising copy and the number of words E of all identical text segments can be counted, and the value F of E / D can be calculated. If F is too small, it means that the identified information point is too small, and the length of the advertising copy is too large. At this time, there is a high probability that some text segments are not completely the same as the preset text segments. At this time, it is necessary to identify the feature text that has not appeared. In order to achieve the purpose of this judgment, a threshold value G related to F can be introduced. If F is less than G, the next step of vector matching is entered to determine the target preset text segment that matches the matching text segment among all preset text segments based on the similarity information between the matching text vector and the preset text vector.

[0189] It can be seen that by implementing this optional implementation method, it is possible to calculate the proportion information of all identical text fragments in the target advertising copy, and determine whether the proportion information is lower than the proportion threshold, so as to determine whether the feature text obtained by character matching is sufficient to represent the feature information in the advertising copy, and perform a vector matching operation when it is determined that it cannot be represented, so as to facilitate the subsequent more accurate extraction of the feature text fragments in the advertising copy, and provide an accurate data basis for the subsequent identification or generation of advertising copy feature information.

[0190] As an optional implementation, Figure 4 As shown, the device also includes:

[0191] The third determination module 305 is used to determine the same text segment as a matching text segment, and determine the feature type of the same text segment as the feature type corresponding to the corresponding preset text segment.

[0192] It can be seen that by implementing this optional implementation, the same text segment matched by characters can also be confirmed as the characteristic text in the copy, providing an accurate data basis for the subsequent recognition or generation of characteristic information of the advertising copy.

[0193] As an optional implementation, Figure 4 As shown, the device also includes a blacklist determination module 306, which is used to perform the following steps:

[0194] Determine the blacklist text data corresponding to each feature type;

[0195] For all matching text segments corresponding to any feature type in the target advertising copy, a feature association cancellation operation is performed for the matching text segments in the blacklist text data corresponding to the feature type; the feature association cancellation operation is used to cancel the association between the matching text segment and its corresponding feature type.

[0196] Optionally, the blacklist text data corresponding to each feature type can be determined by manually extracting the preset text segments that cannot accurately represent the feature type from all preset text segments corresponding to a certain feature type obtained by screening the promotional copy of the network platform, and determining them as the blacklist text data corresponding to the feature type. Among them, the preset text segment that cannot accurately represent the feature type, that is, the text segment in the blacklist text data corresponding to the feature type, can be a text segment that has a high similarity with other text segments that can accurately represent the feature type but is easy to cause ambiguity.

[0197] Specifically, the feature association cancellation operation is used to cancel the association between the matching text segment and its corresponding feature type. Specifically, the feature association cancellation operation can also be used to delete the matching text segment in the blacklist text data corresponding to the feature type or cancel its identification as a matching text segment. Optionally, the blacklist text data can also correspond to multiple feature types at the same time. For example, the blacklist text data can correspond to two feature types. When matching, the matching text segment corresponding to the two feature types will be used to match the blacklist text data corresponding to the two feature types at the same time.

[0198] It can be seen that by implementing this optional implementation method, it is possible to perform a feature association cancellation operation on the matching text fragments in the blacklist text data in the target advertising copy, thereby canceling the feature association of the matching text fragments that cannot accurately represent the corresponding feature type, so as to avoid subsequent distortion of the information related to the feature type of the copy, which is conducive to improving the accuracy of feature information extraction from the advertising copy.

[0199] Embodiment 4

[0200] See also Figure 5 , Figure 5 FIG. 1 is a schematic diagram of the structure of another device for extracting features of advertisement text disclosed in an embodiment of the present invention. Figure 5 As shown, the device may include:

[0201] A memory 401 storing executable program codes;

[0202] a processor 402 coupled to the memory 401;

[0203] The processor 402 calls the executable program code stored in the memory 401 to execute part or all of the steps in the feature extraction method of the advertisement copy disclosed in the first or second embodiment of the present invention.

[0204] Embodiment 5

[0205] An embodiment of the present invention discloses a computer storage medium storing computer instructions. When the computer instructions are called, they are used to execute part or all of the steps in the feature extraction method of the advertisement copy disclosed in the first or second embodiment of the present invention.

[0206] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0207] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution can be essentially or partly contributed to the prior art in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0208] Finally, it should be noted that the method and device for extracting features from advertising copy disclosed in the embodiments of the present invention disclose only preferred embodiments of the present invention, and are only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting features from advertising copy, characterized in that: The method comprises: Determine a plurality of preset text segments corresponding to a plurality of feature types; the feature types include at least one of a category of a promotional object, a component of a promotional object, an efficacy of a promotional object, a brand of a promotional object, and promotional marketing terms; Determining a target advertising copy; the target advertising copy includes a plurality of text segments; Matching the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types; The step of matching the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types includes: Determining a preset text vector corresponding to each of the preset text segments; Determine a matching text segment from all the text segments in the target advertisement copy, and determine a matching text vector of the matching text segment; Determining a target preset text segment that matches the matching text segment among all the preset text segments according to similarity information between the matching text vector and the preset text vector; Determining the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment; The step of determining a matching text segment from all the text segments in the target advertisement copy includes: Determine a part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule; the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments; Determine the text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy as matching text segments; The step of determining the part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule includes: Performing part-of-speech tagging on each of the preset text segments according to a part-of-speech tagging algorithm to obtain multiple part-of-speech combinations; Calculating the number of occurrences of each of the part-of-speech combinations, and retaining the part-of-speech combinations whose number of occurrences is higher than a preset number threshold, to obtain a set of part-of-speech combinations; The step of determining the text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy as matching text segments includes: According to the part-of-speech tagging algorithm, each of the text segments in the target advertisement copy is tagged with a part-of-speech to obtain a plurality of corresponding text part-of-speech combinations; Match any of the text part-of-speech combinations with all the part-of-speech combinations in the part-of-speech combination set, and when the matching result shows that there is a part-of-speech combination in the part-of-speech combination set that is identical to the text part-of-speech combination, determine the text segment corresponding to the text part-of-speech combination as the matching text segment.

2. The feature extraction method of advertisement copy according to claim 1, characterized in that: The step of determining a target preset text segment that matches the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector includes: Based on a clustering algorithm, clustering calculation is performed on all the preset text vectors to obtain multiple preset text vector sets; Determine, according to all the preset text vectors in each of the preset text vector sets, a set vector corresponding to each of the preset text vector sets; Calculating similarity information between the matching text vector and a set vector corresponding to any one of the preset text vector sets, and determining whether the similarity information is greater than a first similarity threshold; If the judgment result is yes, calculating the similarity information between the matching text vector and all the preset text vectors in the preset text vector set; The preset text segment corresponding to the preset text vector having the highest similarity information with the matching text vector in the preset text vector set and / or higher than a second similarity threshold is determined as the target preset text segment.

3. The feature extraction method of advertisement copy according to claim 1 or 2, characterized in that: The similarity information includes at least one of cosine similarity information, Euclidean distance similarity information, Jaccard distance similarity information, edit distance similarity information, Chebyshev distance similarity information, Hamming distance similarity information, Mahalanobis distance similarity information, Manhattan distance similarity information and Minkowski distance similarity information.

4. The method for extracting features from advertisement text according to claim 1, characterized in that: Before determining a target preset text segment that matches the matching text segment among all the preset text segments based on the similarity information between the matching text vector and the preset text vector, the method further includes: Performing character matching on the target advertisement copy and the plurality of preset text segments to obtain a text segment in the target advertisement copy that is identical to any of the preset text segments; Calculating the proportion information of all the same text segments in the target advertisement copy, and determining whether the proportion information is lower than a proportion threshold; If the judgment result is yes, the step of determining a target preset text segment that matches the matching text segment among all the preset text segments according to the similarity information between the matching text vector and the preset text vector is performed; And, the method further comprises: The same text segment is determined as the matching text segment, and the feature type of the same text segment is determined as the feature type corresponding to the corresponding preset text segment.

5. The feature extraction method of advertisement copy according to claim 1, characterized in that: The method further comprises: Determine the blacklist text data corresponding to each of the feature types; For all matching text segments corresponding to any of the feature types in the target advertising copy, a feature association cancellation operation is performed on the matching text segments in the blacklist text data corresponding to the feature type; the feature association cancellation operation is used to cancel the association between the matching text segment and its corresponding feature type.

6. A feature extraction device for advertising copy, characterized in that: The device comprises: The first determination module is used to determine a plurality of preset text segments corresponding to a plurality of feature types; the feature types include at least one of the following: the category of the promotion object, the composition of the promotion object, the efficacy of the promotion object, the brand of the promotion object, and the promotional marketing terms; A second determination module is used to determine a target advertisement copy; the target advertisement copy includes a plurality of text segments; A matching module, configured to match the multiple text segments in the target advertisement copy with the multiple preset text segments to obtain a matching text segment in the target advertisement copy corresponding to any of the feature types; Wherein, the matching module includes: A first determining unit, configured to determine a preset text vector corresponding to each of the preset text segments; A second determining unit, configured to determine a matching text segment from all the text segments in the target advertisement copy, and determine a matching text vector of the matching text segment; A third determining unit, configured to determine a target preset text segment that matches the matching text segment among all the preset text segments according to similarity information between the matching text vector and the preset text vector; A fourth determining unit, configured to determine the feature type corresponding to the matching text segment as the feature type corresponding to the target preset text segment; The specific manner in which the second determination unit determines the matching text segment from all the text segments in the target advertisement copy includes: Determine a part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule; the part-of-speech combination rule is used to represent the part-of-speech combination regularity of the text in the plurality of preset text segments; Determine the text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy as matching text segments; The specific manner in which the second determining unit determines the part-of-speech combination rule according to the plurality of preset text segments and the part-of-speech tagging rule includes: Performing part-of-speech tagging on each of the preset text segments according to a part-of-speech tagging algorithm to obtain multiple part-of-speech combinations; Calculating the number of occurrences of each of the part-of-speech combinations, and retaining the part-of-speech combinations whose number of occurrences is higher than a preset number threshold, to obtain a set of part-of-speech combinations; The specific manner in which the second determination unit determines the text segments that meet the part-of-speech combination rule among all the text segments in the target advertisement copy as matching text segments includes: According to the part-of-speech tagging algorithm, each of the text segments in the target advertisement copy is tagged with a part-of-speech to obtain a plurality of corresponding text part-of-speech combinations; Match any of the text part-of-speech combinations with all the part-of-speech combinations in the part-of-speech combination set, and when the matching result shows that there is a part-of-speech combination in the part-of-speech combination set that is identical to the text part-of-speech combination, determine the text segment corresponding to the text part-of-speech combination as the matching text segment.

7. A feature extraction device for advertising copy, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the feature extraction method for the advertisement copy according to any one of claims 1-5.

Images