An Adaptive Product Classification Method Based on Large Model AI Recognition

Abstract

The present invention relates to the field of label information analysis, and particularly to an adaptive commodity classification method based on large model AI recognition, including determining the text reference status of each relevant analysis text according to the text coincidence parameter and the evaluation reference index; determining an effective analysis strategy according to the text reference status of each relevant analysis text; when performing phrase reference analysis on the text analysis set, determining the text division method according to the evaluation reference index and the relevant evaluation coefficient; when performing effective evaluation analysis on the relevant evaluation keywords, determining the effective evaluation phrases according to the text correlation coefficient and the proportion of key users, and determining the evaluation key coefficient of each effective evaluation phrase according to the text relevance and the user reference coefficient; under the condition that the evaluation extraction is completed, determining the label update strategy according to the effective key parameter and the reference key coefficient to determine the warning update label, and the present invention improves the reliability of the optimization result of the label data.

Description

Technical Field

[0001] The present invention relates to the field of label information analysis, and particularly to an adaptive commodity classification method based on large model AI recognition. Background Art

[0002] Fine classification of commodities can effectively improve the communication efficiency between consumers and merchants. In the process of classifying commodities by e-commerce platforms, it is usually based on the commodity label information provided by merchants. However, in order to increase the probability of commodity retrieval or due to insufficient understanding of commodity classification by merchants, it is easy to lead to low-quality commodity label information, resulting in low accuracy of commodity classification. Optimizing the label information through the information content of various comments on commodities can effectively improve the accuracy of commodity classification results. However, the comment content often has problems such as low content quality and overlapping comments. Therefore, how to ensure the effectiveness of the comment content used to optimize the label information is an urgent problem to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN117973392A discloses a data automatic acquisition method based on web page label analysis. The method includes the following steps: extracting inter-cluster structure features from the obtained target acquisition web page data to obtain initial DOM tree structure feature data; performing page dynamic loading monitoring according to the initial DOM tree structure feature data to generate page dynamic loading data; performing dynamic label combination processing according to the page dynamic loading data to obtain dynamic label combination data; performing brand reputation label recognition according to the dynamic label combination data to generate brand reputation label data; performing adaptive acquisition optimization according to the brand reputation label data to generate an adaptive page acquisition strategy; performing multi-modal brand data acquisition according to the adaptive page acquisition strategy and performing commodity reputation prediction to obtain commodity reputation prediction data. However, the above technical solution has the following problems: it fails to screen out effective information from the comment information based on which the label data is optimized, resulting in low reliability of the optimization result of the label data. Summary of the Invention

[0004] Therefore, the present invention provides an adaptive commodity classification method based on large model AI recognition to overcome the problem in the prior art that effective information screening is not carried out on the comment information based on which the label data is optimized, resulting in low reliability of the optimization result of the label data, and further resulting in low accuracy of the commodity classification result.

[0005] To achieve the above object, the present invention provides an adaptive commodity classification method based on large model AI recognition, including:

[0006] Obtaining relevant analysis texts of target classification commodities, and determining the text reference status of each relevant analysis text according to the text coincidence parameter and the evaluation reference index;

[0007] Determine an effective analysis strategy based on the text reference status of each relevant analysis text. The effective analysis strategy includes performing phrase reference analysis on the text analysis set and, for relevant evaluation keywords, performing effective evaluation analysis;

[0008] When performing phrase reference analysis on the text analysis set, determine the text division method according to the evaluation reference index and the relevant evaluation coefficient, and determine the setting method of the evaluation key coefficient according to the text division method. The text division method is to determine the text analysis set according to the evaluation relevance or to determine the text analysis set according to the evaluation reference tendency coefficient and the reference relevant evaluation coefficient;

[0009] When performing effective evaluation analysis on relevant evaluation keywords, determine the effective evaluation phrases according to the text correlation coefficient and the proportion of key users, and determine the evaluation key coefficient of each effective evaluation phrase according to the text relevance and the user reference coefficient;

[0010] Under the condition that the evaluation extraction is completed, determine the label update strategy according to the effective key parameters and the reference key coefficient to determine the warning update label. The label update strategy is to determine the evaluation phrase set of each label keyword according to the difference value of the tendency coefficient or to determine the update optimization coefficient of each label keyword according to the evaluation tendency coefficient and the evaluation key coefficient of the effective matching phrase.

[0011] Furthermore, determine the text reference status of each relevant analysis text according to the text coincidence parameter and the evaluation reference index;

[0012] If the text coincidence parameter of a relevant analysis text is greater than the preset text coincidence parameter or the evaluation reference index is less than or equal to the preset evaluation reference index, then determine that this relevant analysis text is in the first preset text reference status;

[0013] If the text coincidence parameter of a relevant analysis text is less than or equal to the preset text coincidence parameter and the evaluation reference index is greater than the preset evaluation reference index, then determine that this relevant analysis text is in the second preset text reference status.

[0014] Furthermore, determine an effective analysis strategy according to the text reference status of each relevant analysis text;

[0015] If a relevant analysis text is in the first preset text reference status, then perform phrase reference analysis on the text analysis set;

[0016] If a relevant analysis text is in the second preset text reference status, then perform effective evaluation analysis on relevant evaluation keywords.

[0017] Furthermore, the division method of the text analysis set is determined according to the text division coefficient;

[0018] If the text division coefficient of a relevant analysis text is greater than the preset text division coefficient, determine the text analysis set according to the evaluation relevance;

[0019] If the text division coefficient of a relevant analysis text is less than or equal to the preset text division coefficient, determine the text analysis set according to the evaluation reference tendency coefficient and the reference correlation evaluation coefficient;

[0020] The text division coefficient is determined according to the evaluation reference index and the relevant evaluation coefficient.

[0021] Further, for a single text analysis set, the phrase reference analysis process includes:

[0022] Determine the effective evaluation phrases according to the text correlation coefficient and the relevant evaluation coefficient;

[0023] Determine the setting method of the evaluation key coefficient of the effective evaluation phrases according to the division method of the text analysis set;

[0024] If the text analysis set is determined according to the evaluation relevance, set the evaluation key coefficient for each effective evaluation phrase of the text analysis set according to the reference coincidence coefficient, and the evaluation key coefficient has a negative correlation with the reference coincidence coefficient;

[0025] If the text analysis set is determined according to the evaluation reference tendency coefficient and the reference correlation coefficient, set the evaluation key coefficient for each effective evaluation phrase of the text analysis set according to the set abnormal tendency parameter and the set reference correlation coefficient, and the evaluation key coefficient has a positive correlation with the set evaluation index.

[0026] Further, extract relevant evaluation phrases for each relevant analysis text in the second preset text reference state, and determine the evaluation effective coefficient of the evaluation phrases according to the text correlation coefficient and the key user proportion;

[0027] Record the evaluation phrases with the evaluation effective coefficient greater than the preset evaluation effective coefficient as effective evaluation phrases, and conduct effective evaluation analysis on the evaluation key coefficient of each effective evaluation phrase.

[0028] Further, the effective evaluation analysis process for a single effective evaluation phrase includes:

[0029] Detect the user reference coefficient of the relevant analysis text of the effective evaluation phrase, and the user reference coefficient is determined according to the user evaluation quality parameter and the historical category coincidence parameter;

[0030] Determine the evaluation key coefficient of the effective evaluation phrase according to the text relevance and the user reference coefficient;

[0031] The described evaluation key coefficients are positively correlated with the text relevance and the user reference coefficient respectively.

[0032] Further, under the condition that the evaluation extraction is completed, a label update strategy is determined according to the effective key parameters and the reference key coefficient;

[0033] If the effective key parameter is greater than the preset effective key parameter or the reference key coefficient is greater than the preset reference key coefficient, the evaluation phrase set of each label keyword is determined according to the difference value of the tendency coefficient;

[0034] If the effective key parameter is less than or equal to the preset effective key parameter and the reference key coefficient is less than or equal to the preset reference key coefficient, the update and optimization coefficient of each label keyword is determined according to the evaluation tendency coefficient and the evaluation key coefficient of the effective matching phrase;

[0035] The described evaluation extraction completion condition is that the effective key analysis of each relevant analysis text of the target classification commodity is completed.

[0036] Further, for a single label keyword, the update and optimization coefficient of each label keyword is determined according to the tendency conflict coefficient and the set reference key coefficient of each evaluation phrase set;

[0037] The described update and optimization coefficient is positively correlated with the tendency conflict coefficient and the set reference key coefficient respectively;

[0038] The difference value of the tendency coefficient of any evaluation phrase set is less than the preset difference value of the tendency coefficient.

[0039] Further, when determining the update and optimization coefficient of each label keyword according to the evaluation tendency coefficient and the evaluation key coefficient of the effective matching phrase, the update and optimization coefficient is determined according to the reference tendency parameter and the relevant key coefficient;

[0040] The described update and optimization coefficient is positively correlated with the relevant matching parameter and the relevant key coefficient respectively;

[0041] An early warning update label is determined according to the update and optimization coefficient of each label keyword, and the label keyword with the update and optimization coefficient greater than the preset update and optimization coefficient is recorded as the early warning update label.

[0042] Compared with the prior art, the beneficial effect of the present invention is that in the technical solution of the present invention, a targeted effective analysis strategy is determined according to the text coincidence parameter and the evaluation reference index of the relevant analysis text of the target classification commodity, so that the method for extracting the effective evaluation phrase is more in line with the actual text situation, so as to screen the evaluation content and improve the effectiveness of the text used as the basis for label optimization. The present invention improves the reliability of the optimization result of the label data.

[0043] Furthermore, in the present invention, the text reference status of each relevant analysis text is determined based on the text overlap parameter and the evaluation reference index, so as to distinguish the proportion of the content available for subsequent label optimization and the situation of whether there are duplicate comments in different relevant analysis texts, which is used to characterize the reference quality of the evaluation content included in the relevant analysis text. The subsequent effective analysis strategy is determined according to the text reference status, avoiding the interference caused by duplicate and low-reference-value evaluation phrases to the label optimization process, and improving the effectiveness of the text used as the basis for label optimization.

[0044] Furthermore, in the present invention, the text analysis set is determined according to the text division coefficient, and the setting method of the evaluation key coefficient of the effective evaluation phrases extracted for different text analysis sets is determined. The text division coefficient is determined through the evaluation reference index and the relevant evaluation coefficient to determine the situation of the content related to the label information and the target classification commodity included in the text. Based on this, the relevant analysis texts in the first preset text reference status are combined, and the extraction of effective evaluation phrases is completed, improving the efficiency of the effective analysis process while reducing the interference of duplicate content to the label optimization process.

[0045] Furthermore, in the present invention, relevant evaluation phrases are extracted for each relevant analysis text in the second preset text reference status, and effective evaluation analysis is performed on the evaluation key coefficients of each effective evaluation phrase. During the effective evaluation analysis process, for the content with higher text quality, the historical evaluation situation of its users is analyzed, and based on this, the evaluation key coefficients of the effective evaluation phrases are set. The present invention improves the reliability of the optimization result of the label data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the adaptive commodity classification method based on large model AI recognition of the present invention;

[0047] Figure 2 It is a flowchart of the present invention for determining the text reference status of each relevant analysis text according to the text overlap parameter and the evaluation reference index;

[0048] Figure 3 It is a flowchart of the present invention for determining the effective analysis strategy according to the text reference status of each relevant analysis text;

[0049] Figure 4 It is a flowchart of the present invention for determining the division method of the text analysis set according to the text division coefficient. DETAILED DESCRIPTION OF THE INVENTION

[0050] To make the objectives and advantages of the present invention more clear and understandable, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0052] It should be noted that in the description of the present invention, the terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0053] In addition, it should also be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] Please refer to Figures 1 to 4 As shown, the present invention provides an adaptive commodity classification method based on large model AI recognition, including:

[0055] Obtain relevant analysis texts of target classification commodities, and determine the text reference status of each relevant analysis text according to the text coincidence parameter and the evaluation reference index;

[0056] Determine an effective analysis strategy according to the text reference status of each relevant analysis text. The effective analysis strategy includes performing phrase reference analysis on the text analysis set and performing effective evaluation analysis on relevant evaluation keywords;

[0057] When performing phrase reference analysis on the text analysis set, determine the text division method according to the evaluation reference index and the relevant evaluation coefficient, and determine the setting method of the evaluation key coefficient according to the text division method. The text division method is to determine the text analysis set according to the evaluation relevance, or to determine the text analysis set according to the evaluation reference tendency coefficient and the reference relevant evaluation coefficient;

[0058] When conducting effective evaluation and analysis on relevant evaluation keywords, effective evaluation phrases are determined according to the text correlation coefficient and the proportion of key users, and the evaluation key coefficients of each effective evaluation phrase are determined according to the text relevance and the user reference coefficient;

[0059] Under the condition that the evaluation extraction is completed, a label update strategy is determined according to the effective key parameters and the reference key coefficient to determine the warning update label. The label update strategy is to determine the evaluation phrase set of each label keyword according to the difference value of the tendency coefficient, or to determine the update optimization coefficient of each label keyword according to the evaluation tendency coefficient and the evaluation key coefficient of the effective matching phrase.

[0060] Among them, the present invention is used to update and optimize the product label information involved in the product classification process. The product for which the product label information is updated and optimized once is recorded as the target classification product, and the relevant evaluation text of the target classification product is obtained. The relevant evaluation text is the evaluation content of the purchasing user of the target classification product for the target classification product. How to obtain the relevant evaluation text is easy for those skilled in the art to understand. The text published by the purchasing user on the product purchase platform used by him / her can be recorded as the relevant evaluation text, which will not be elaborated here. Each product in the present invention is provided with a preset category, and the categories of the preset category include but are not limited to: household appliances, digital devices, clothing, and beauty products.

[0061] The relevant review texts of the target classified products are recorded as relevant analysis texts, and the product description texts of the target classified products themselves are recorded as label texts. In the present invention, the trained phrase recognition model is used to perform AI recognition on the relevant analysis texts and label texts to obtain label keywords and evaluation phrases. The evaluation tendency coefficients of each evaluation phrase are determined according to the product evaluation scores corresponding to the relevant analysis texts. For a single evaluation phrase, the evaluation tendency coefficient is the average value of the product evaluation scores of the relevant analysis texts in which the evaluation phrase exists. The product evaluation score is the evaluation score given to the product when the relevant analysis text is published. How to obtain the product evaluation score is easy for those skilled in the art to understand and will not be elaborated here. The process of performing AI recognition on the label keywords and evaluation phrases is based on the trained phrase recognition model. How to train the phrase recognition model is easy for those skilled in the art to understand. The training process of the phrase recognition model includes: pre-training the phrase recognition model using a large amount of unsupervised data, learning the general representation of the language through the task of predicting the next word, preparing a dataset of labeled analysis texts and label texts, and these data need to be cleaned and labeled. On the basis of the pre-trained phrase recognition model, fine-tuning is carried out. The initialized model parameters are the same as those of the pre-trained phrase recognition model. The analysis text is input into the model, and the model outputs the predicted label keywords and evaluation phrases. By calculating the loss (such as cross-entropy loss) between the prediction result and the true label, the backpropagation algorithm is used to adjust the model parameters to minimize the loss. At the same time, hyperparameters such as the learning rate and batch size need to be adjusted to optimize the training effect of the model. Users can set the training process of the phrase recognition model according to actual needs, which is easy for those skilled in the art to understand and will not be elaborated here. The content of the evaluation phrases in the present invention includes but is not limited to: high power consumption, loud noise, and allergies. The content of the label keywords in the present invention includes but is not limited to: energy consumption, quietness, and mildness.

[0062] In the present invention, there are several label optimization records. Any one of the label optimization records records the number of relevant analysis texts of the repeated continuous paragraphs, the number of words in each repeated continuous paragraph, the reference evaluation coefficient, the text coincidence parameter, the evaluation reference index, the text division coefficient, the evaluation relevance, the set difference coefficient, the effective evaluation parameter, the evaluation effectiveness coefficient, the user evaluation quality parameter, the effective key parameter, the reference key coefficient, the update optimization coefficient, and the tendency coefficient difference value during the update and optimization process of the label information for a target classified product at least once. And each label optimization record corresponds to a qualified mark, and the qualified mark records whether the accuracy of the label information optimization result meets the user's needs. It can be understood that users can determine whether the accuracy of the label information optimization result meets the requirements according to the self-set indicators. For example, the self-set indicators can be but are not limited to the product recommendation accuracy, and the product recommendation accuracy is the evaluation satisfaction degree of the users collected in the form of a questionnaire after the label information optimization is completed.

[0063] Specifically, the text reference status of each relevant analysis text is determined according to the text coincidence parameter and the evaluation reference index;

[0064] If the text coincidence parameter of a relevant analysis text is greater than the preset text coincidence parameter or the evaluation reference index is less than or equal to the preset evaluation reference index, it is determined that the relevant analysis text is in the first preset text reference status;

[0065] If the text coincidence parameter of a relevant analysis text is less than or equal to the preset text coincidence parameter and the evaluation reference index is greater than the preset evaluation reference index, it is determined that the relevant analysis text is in the second preset text reference status.

[0066] Among them, for a single relevant analysis text, the text coincidence parameter = the sum of the word counts of each repeated continuous paragraph / the total word count of the relevant analysis text. For a continuous paragraph existing in the relevant analysis text, if the number of relevant analysis texts containing this continuous paragraph is greater than the preset repeated text number, this continuous paragraph is recorded as a repeated continuous paragraph. The continuous paragraph is a text paragraph whose contained word count is greater than the preset continuous word count. The text paragraph is composed of one or more sentences and is separated according to line breaks, spaces or punctuation marks for the relevant analysis text to obtain several text paragraphs. The values of the preset repeated text number and the preset continuous word count can be determined by the user according to the actual working scenario. For example, the user can set according to the tag optimization record. A method for obtaining the value of the preset repeated text number is provided, and the minimum value of the number of relevant analysis texts containing each repeated continuous paragraph in the tag optimization record that meets the user's accuracy requirement for the tag information optimization result is recorded as the preset repeated text number. A method for obtaining the value of the preset continuous word count is provided, and the average value of the word counts of each repeated continuous paragraph in the tag optimization record that meets the user's accuracy requirement for the tag information optimization result is recorded as the preset continuous word count;

[0067] The evaluation reference index is the number of reference evaluation phrases contained in the relevant analysis text. The reference evaluation phrase is an evaluation phrase whose reference evaluation coefficient is greater than the preset reference evaluation coefficient. For a single evaluation phrase, the reference evaluation coefficient is the maximum value of the number of times the evaluation phrase is an effective matching phrase of each label keyword of the target classification product. The value of the preset reference evaluation coefficient can be determined by the user according to the actual working scenario. For example, the user can set according to the tag optimization record. The higher the user's accuracy requirement for the tag information optimization result, the greater the value of the preset reference evaluation coefficient. A method for obtaining the value of the preset reference evaluation coefficient is provided, and the minimum value of the reference evaluation coefficients of each reference evaluation phrase in the tag optimization record that meets the user's accuracy requirement for the tag information optimization result is recorded as the preset reference evaluation coefficient;

[0068] The values of the preset text coincidence parameter and the preset evaluation reference index can be determined by the user according to the actual working scenario. For example, the user can set according to the label optimization record. The higher the user's accuracy requirement for the optimization result of the label information, the smaller the value of the preset text coincidence parameter and the larger the value of the preset evaluation reference index. A method for obtaining the value of the preset text coincidence parameter is provided. The minimum value of the text coincidence parameter of the relevant analysis text in the first preset text reference state in the label optimization record that meets the user's accuracy requirement for the optimization result of the label information is recorded as the preset text coincidence parameter. A method for obtaining the value of the preset evaluation reference index is provided. The minimum value of the evaluation reference index of the relevant analysis text in the first preset text reference state in the label optimization record that meets the user's accuracy requirement for the optimization result of the label information is recorded as the preset evaluation reference index.

[0069] Specifically, an effective analysis strategy is determined according to the text reference state of each relevant analysis text;

[0070] If a relevant analysis text is in the first preset text reference state, a phrase reference analysis is performed on the text analysis set;

[0071] If a relevant analysis text is in the second preset text reference state, an effective evaluation analysis is performed on the relevant evaluation keywords.

[0072] Specifically, the division method of the text analysis set is determined according to the text division coefficient;

[0073] If the text division coefficient of a relevant analysis text is greater than the preset text division coefficient, the text analysis set is determined according to the evaluation relevance;

[0074] If the text division coefficient of a relevant analysis text is less than or equal to the preset text division coefficient, the text analysis set is determined according to the evaluation reference tendency coefficient and the reference correlation coefficient;

[0075] The text division coefficient is determined according to the evaluation reference index and the relevant evaluation coefficient.

[0076] Among them, for a single relevant analysis text, the text division coefficient = ln(evaluation reference index × reference correlation evaluation coefficient). The reference correlation evaluation coefficient is the average value of the correlation evaluation coefficients of each evaluation phrase included in this relevant analysis text. For a single evaluation phrase, the correlation evaluation coefficient is the product of the category correlation quantity and the category correlation frequency. The category correlation quantity is the number of products of the relevant category of this evaluation phrase. The products in the relevant analysis text that have this evaluation phrase and are of the same preset category as the target classification products are recorded as the relevant category products of this evaluation phrase. The category correlation frequency is the average value of the number of relevant analysis texts in which each relevant category product has this evaluation phrase. The value of the preset text division coefficient can be determined by the user according to the actual working scenario. For example, the user can set it according to the label optimization record, and a method for obtaining the value of the preset text division coefficient is provided. The label optimization record for determining the text analysis set according to the evaluation relevance is recorded as the division reference record, and the minimum value of the text division coefficients of the relevant analysis texts in the division reference record that meets the user's accuracy requirement for the label information optimization result is recorded as the preset text division coefficient;

[0077] The evaluation relevance of any text analysis set determined according to the evaluation relevance is greater than the preset evaluation relevance. The evaluation relevance = the number of overlapping evaluation phrases in the text analysis set / the number of evaluation phrases included in the text analysis set. For a single evaluation phrase, the relevant analysis texts that have this evaluation phrase are recorded as associated texts. If the number of associated texts of this evaluation phrase in a text analysis set that includes this evaluation phrase is greater than 1, then this evaluation phrase is recorded as the overlapping evaluation phrase of this text analysis set. The value of the preset evaluation relevance can be determined by the user according to the actual working scenario. For example, the user can set it according to the label optimization record. The higher the user's accuracy requirement for the label information optimization result, the larger the value of the preset evaluation relevance. A method for obtaining the value of the preset evaluation relevance is provided. The average value of the evaluation relevances of each text analysis set in the division reference record that meets the user's accuracy requirement for the label information optimization result is recorded as the preset evaluation relevance;

[0078] The set difference coefficient of any text analysis set determined according to the evaluation reference tendency coefficient and the reference correlation coefficient is less than the preset set difference coefficient. For a single text analysis set, the set difference coefficient is the sum of the products of the tendency coefficient difference value and the correlation coefficient difference value and their corresponding difference influence coefficients. The tendency coefficient difference value , where n is the number of relevant analysis texts included in this text analysis set, is the reference tendency coefficient of the i-th relevant analysis text in this text analysis set, is the average value of the reference tendency coefficients of each relevant analysis text in the text analysis set. For a single relevant analysis text, the reference tendency coefficient is the average value of the evaluation tendency coefficients of each comment phrase included in the relevant analysis text, and the correlation coefficient difference value , is the reference correlation coefficient of the i-th relevant analysis text in the text analysis set, is the average value of the reference correlation coefficients of each relevant analysis text in the text analysis set. The reference correlation coefficient is the average value of the relevant evaluation coefficients of each comment phrase included in the relevant analysis text. For the values of the difference impact coefficients corresponding to the tendency coefficient difference value and the correlation coefficient difference value, the user can determine them according to the actual working scenario. Provide a value of the difference impact coefficient corresponding to the tendency coefficient difference value, and the value of the difference impact coefficient corresponding to the tendency coefficient difference value is 0.5. Provide a value of the difference impact coefficient corresponding to the correlation coefficient difference value, and the value of the difference impact coefficient corresponding to the correlation coefficient difference value is 0.5;

[0079] The value of the preset set difference coefficient can be determined by the user according to the actual working scenario. For example, the user can set it according to the label optimization record. The higher the user's requirement for the accuracy of the label information optimization result, the smaller the value of the preset set difference coefficient. Provide a method for determining the value of the preset set difference coefficient. Record the label optimization record of the text analysis set determined according to the evaluation tendency coefficient and the relevant evaluation coefficient as the reference difference record, and record the maximum value of the set difference coefficient of the text analysis set in the reference difference record that meets the user's requirement for the accuracy of the label information optimization result as the preset set difference coefficient.

[0080] Specifically, for a single text analysis set, the phrase reference analysis process includes:

[0081] Determine the effective evaluation phrases according to the text correlation coefficient and the relevant evaluation coefficient;

[0082] Determine the setting method of the evaluation key coefficient of the effective evaluation phrases according to the division method of the text analysis set;

[0083] If the text analysis set is determined according to the evaluation relevance, set the evaluation key coefficients of the effective evaluation phrases of the text analysis set according to the reference coincidence coefficient, and the evaluation key coefficient and the reference coincidence coefficient are negatively correlated;

[0084] If the text analysis set is determined according to the reference tendency coefficient and the reference correlation coefficient, set the evaluation key coefficients of the effective evaluation phrases of the text analysis set according to the set abnormal tendency parameter and the set reference correlation coefficient, and the evaluation key coefficient is positively correlated with the set evaluation index.

[0085] Among them, for a single text analysis set, the effective evaluation phrase is an evaluation phrase whose effective evaluation parameter is greater than a preset effective evaluation parameter. For a single evaluation phrase, the effective evaluation parameter is the product of a text correlation coefficient and a relevant evaluation coefficient. The text correlation coefficient is the number of relevant analysis texts containing this evaluation phrase in this text analysis set. The value of the preset effective evaluation parameter can be determined by the user according to the actual working scenario. For example, the user can set it according to the label optimization record. The higher the user's accuracy requirement for the label information optimization result, the larger the value of the preset effective evaluation parameter. A method for obtaining the value of the preset effective evaluation parameter is provided. The minimum value of the effective evaluation parameters of the effective evaluation phrases in the division reference records that meet the user's accuracy requirement for the label information optimization result is recorded as the preset effective evaluation parameter;

[0086] If a text analysis set is determined according to the evaluation relevance, the evaluation key coefficients of the effective evaluation phrases included in this text analysis set are determined according to the reference coincidence coefficient, and the reference coincidence coefficient is the average value of the text coincidence parameters of the relevant analysis texts in this text analysis set; if a text analysis set is determined according to the evaluation sentiment coefficient and the relevant evaluation coefficient, the evaluation key coefficients of the effective evaluation phrases included in this text analysis set are determined according to the set abnormal sentiment parameter and the set reference correlation coefficient. The set evaluation index is determined according to the set abnormal sentiment parameter and the set reference correlation coefficient. The set evaluation index = the set reference correlation coefficient × the index weight coefficient corresponding to the set reference correlation coefficient - the set abnormal tendency parameter × the index weight coefficient corresponding to the set abnormal tendency parameter. For a single text analysis set, the set reference correlation coefficient is the average value of the reference correlation coefficients of the relevant analysis texts in this text analysis set, and the set abnormal tendency parameter is the absolute value of the difference between the average value of the reference tendency coefficients of the relevant analysis texts in this text analysis set and the average value of the reference tendency coefficients of the relevant analysis texts of the target classification commodity. The values of the index weight coefficients corresponding to the set reference correlation coefficient and the set abnormal tendency parameter can be determined by the user according to the actual working scenario. A method for obtaining the value of the index weight coefficient corresponding to the set reference correlation coefficient is provided as 0.6, and a method for obtaining the value of the index weight coefficient corresponding to the set abnormal tendency parameter is provided as 0.4.

[0087] Specifically, relevant evaluation phrases are extracted for each relevant analysis text in the second preset text reference state, and the evaluation effective coefficient of the evaluation phrase is determined according to the text correlation coefficient and the key user proportion;

[0088] The evaluation phrases with an evaluation effective coefficient greater than the preset evaluation effective coefficient are recorded as effective evaluation phrases, and effective evaluation analysis is performed on the evaluation key coefficients of each effective evaluation phrase.

[0089] Among them, for a single evaluation phrase, the evaluation effectiveness coefficient is the sum of the products of the text correlation coefficient and the proportion of key users and their corresponding evaluation influence coefficients respectively. The proportion of key users = the number of key users / the number of relevant users. The relevant users are the purchasing users corresponding to the relevant analysis text containing the evaluation phrase. The key users are the relevant users whose user evaluation quality parameter is greater than the preset user evaluation quality parameter. The values of the evaluation influence coefficients corresponding to the text correlation coefficient and the proportion of key users can be determined by the user according to the actual working scenario. A method for determining the values of the evaluation influence coefficients corresponding to the text correlation coefficient and the proportion of key users is provided. The value of the evaluation influence coefficient corresponding to the text correlation coefficient is 0.4, and the value of the evaluation influence coefficient corresponding to the proportion of key users is 0.6. For a single relevant user, the user evaluation quality parameter = the number of texts for optimization basis / the number of historical evaluation texts. The historical evaluation texts are the relevant analysis texts published by the relevant user in the label optimization record. The texts for optimization basis are the historical evaluation texts with valid evaluation phrases in the corresponding label optimization record.

[0090] The values of the preset evaluation effectiveness coefficient and the preset user evaluation quality parameter can be determined by the user according to the actual working scenario. For example, the user can set them according to the label optimization record. The higher the user's requirement for the accuracy of the label information optimization result, the larger the value of the preset evaluation effectiveness coefficient and the larger the value of the preset user evaluation quality parameter. A method for determining the value of the preset evaluation effectiveness coefficient is provided. The label optimization record for extracting relevant evaluation phrases from the relevant analysis text in the second preset text reference state is recorded as a valid analysis record. The minimum value of the evaluation effectiveness coefficients of the valid evaluation phrases in the valid analysis records that meet the user's requirement for the accuracy of the label information optimization result is recorded as the preset evaluation effectiveness coefficient. A method for determining the value of the preset user evaluation quality parameter is provided. The average value of the user evaluation quality parameters of each key user in the valid analysis records that meet the user's requirement for the accuracy of the label information optimization result is recorded as the preset user evaluation quality parameter.

[0091] Specifically, the effective evaluation analysis process for a single valid evaluation phrase includes:

[0092] Detect the user reference coefficient of the relevant analysis text of the valid evaluation phrase, where the user reference coefficient is determined according to the user evaluation quality parameter and the historical category coincidence parameter;

[0093] Determine the evaluation key coefficient of the valid evaluation phrase according to the text relevance and the user reference coefficient;

[0094] The evaluation key coefficient has a positive correlation with the text relevance and the user reference coefficient respectively.

[0095] Among them, for a single valid evaluation phrase, its user reference coefficient is the sum of the reference evaluation quality parameter and the historical category overlap parameter. The historical category overlap parameter = the number of overlapping category texts of the relevant users of this valid evaluation phrase / the number of historical evaluation texts of the relevant users of this valid evaluation phrase. If the product category corresponding to a historical evaluation text is the same as the target classification product, then this historical evaluation text is recorded as an overlapping category text. The historical evaluation text is the relevant review text of each product in the label optimization record. The reference evaluation quality parameter is the average value of the user evaluation quality parameters of each relevant user of this valid evaluation phrase. The text relevance is the number of relevant analysis texts containing this valid evaluation phrase. The evaluation key coefficient is the sum of the text relevance and the user reference coefficient.

[0096] Specifically, under the condition that the evaluation extraction is completed, determine the label update strategy according to the valid key parameter and the reference key coefficient.

[0097] If the valid key parameter is greater than the preset valid key parameter or the reference key coefficient is greater than the preset reference key coefficient, then determine the evaluation phrase set of each label keyword according to the difference value of the tendency coefficient.

[0098] If the valid key parameter is less than or equal to the preset valid key parameter and the reference key coefficient is less than or equal to the preset reference key coefficient, then determine the update and optimization coefficient of each label keyword according to the evaluation tendency coefficient and the evaluation key coefficient of the valid matching phrase.

[0099] The condition for the completion of the evaluation extraction is that all relevant analysis texts of the target classification product have completed the valid key analysis.

[0100] Among them, the valid key parameter is the number of determined valid evaluation phrases, and the reference key coefficient is the average value of the evaluation key coefficients of each valid evaluation phrase. The values of the preset valid key parameter and the preset reference key coefficient can be determined by the user according to the actual working scenario. For example, the user can set according to the label optimization record. Provide a method for obtaining the value of the preset valid key parameter. Record the label optimization record that determines the evaluation phrase set of each label keyword according to the difference value of the tendency coefficient as the first update record, and record the minimum value of the valid key parameter in the first update record that meets the user's accuracy requirement for the label information optimization result as the preset valid key parameter. Provide a method for obtaining the value of the preset reference key coefficient. Record the minimum value of the reference key coefficient in the first update record that meets the user's accuracy requirement for the label information optimization result as the preset reference key coefficient. The completion of the valid key analysis of all relevant analysis texts of the target classification product means that the extraction of the valid evaluation phrase is completed and the evaluation key coefficients of each valid evaluation phrase are set.

[0101] Specifically, for a single label keyword, the update and optimization coefficient of each label keyword is determined according to the tendency conflict coefficient and the set reference key coefficient of each evaluation phrase set;

[0102] The update and optimization coefficients are respectively in a positive correlation with the tendency conflict coefficient and the reference key coefficient;

[0103] The difference value of the tendency coefficients of any evaluation phrase set is less than the preset difference value of the tendency coefficients.

[0104] Among them, for a single evaluation phrase set, the difference value of the tendency coefficients , where m is the number of effective matching phrases included in this evaluation phrase set, pj is the evaluation tendency coefficient of the jth effective matching phrase in this evaluation phrase set, is the average value of the evaluation tendency coefficients of the effective matching phrases included in this evaluation phrase set. The value of the preset difference value of the tendency coefficients can be determined by the user according to the actual working scenario. For example, the user can set it according to the label optimization record. A method for obtaining the value of the preset difference value of the tendency coefficients is provided. The minimum value of the difference value of the tendency coefficients of the evaluation phrase set in the first update record that meets the user's accuracy requirements for the label information optimization result is recorded as the preset difference value of the tendency coefficients;

[0105] For a single label keyword, the update and optimization coefficient , where h is the number of evaluation phrase sets of this label keyword, and are respectively the tendency conflict coefficient and the set reference key coefficient of the s-th evaluation phrase set of this label keyword, and are respectively the influence coefficients corresponding to the tendency conflict coefficient and the set reference key coefficient. For a single evaluation phrase set, the set reference key coefficient is the average value of the evaluation key coefficients of the effective matching phrases included in this evaluation phrase set. The tendency conflict coefficient is the absolute value of the difference between the tendency reference value of this evaluation phrase set and the average tendency reference value. The tendency reference value is the average value of the evaluation tendency coefficients of the effective matching phrases included in this evaluation phrase set. The average tendency reference value is the average value of the tendency reference values of the evaluation phrase sets of this label keyword. The values of the influence coefficients corresponding to the tendency conflict coefficient and the set reference key coefficient can be determined by the user according to the actual working scenario. A value for the influence coefficient corresponding to the tendency conflict coefficient is provided. The value of the influence coefficient corresponding to the tendency conflict coefficient is 0.5. A value for the influence coefficient corresponding to the set reference key coefficient is provided. The value of the influence coefficient corresponding to the tendency conflict coefficient is 0.5.

[0106] Specifically, when determining the update and optimization coefficient of each tag keyword according to the evaluation tendency coefficient and evaluation key coefficient of the effective matching phrases, the update and optimization coefficient is determined according to the reference tendency parameter and the relevant key coefficient;

[0107] The update and optimization coefficient is positively correlated with the relevant matching parameter and the relevant key coefficient respectively;

[0108] Determine the warning update tag according to the update and optimization coefficient of each tag keyword, and record the tag keyword with the update and optimization coefficient greater than the preset update and optimization coefficient as the warning update tag.

[0109] Among them, for a single tag keyword, the update and optimization coefficient = ln (reference tendency parameter × relevant key coefficient), the reference tendency parameter is the average value of the evaluation tendency coefficients of the effective matching phrases of this tag keyword, and the relevant key coefficient is the average value of the evaluation key coefficients of the effective matching phrases of this tag keyword. The value of the preset update and optimization coefficient can be determined by the user according to the actual working scenario. For example, the user can set it according to the tag optimization record. The higher the user's accuracy requirement for the optimization result of the tag information, the larger the value of the preset update and optimization coefficient. Provide a method for determining the value of the preset update and optimization coefficient, and record the minimum value of the update and optimization coefficients of each warning update tag in the tag optimization record that meets the user's accuracy requirement for the optimization result of the tag information as the preset update and optimization coefficient. If there is a warning update tag for the target classified commodity, send a tag update requirement to the user, and delete or replace the content of the warning update tag.

[0110] So far, the technical solution of the present invention has been described in combination with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or replacements to the relevant technical features, and the technical solutions after these changes or replacements will all fall within the protection scope of the present invention.

[0111] The above are only the preferred embodiments of the present invention and are not used to limit the present invention; for those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An adaptive commodity classification method based on AI recognition of a large model, characterized in that: include: Obtain relevant analysis texts of target category products, and determine the text reference status of each relevant analysis text according to the text overlap parameter and the evaluation reference index; Determining an effective analysis strategy according to the text reference status of each relevant analysis text, the effective analysis strategy including performing phrase reference analysis on the text analysis set, and performing effective evaluation analysis on relevant evaluation keywords; When performing phrase reference analysis on a text analysis set, a text division method is determined according to an evaluation reference index and a related evaluation coefficient, and a setting method of the evaluation key coefficient is determined according to the text division method, wherein the text division method is to determine a text analysis set according to an evaluation relevance, or to determine a text analysis set according to an evaluation reference tendency coefficient and a reference related evaluation coefficient; When conducting effective evaluation analysis on relevant evaluation keywords, determine effective evaluation phrases based on text relevance coefficients and key user proportions, and determine the evaluation key coefficients of each effective evaluation phrase based on text relevance and user reference coefficients; Under the condition that the evaluation extraction is completed, the label update strategy is determined according to the effective key parameters and the reference key coefficients to determine the warning update label. The label update strategy is to determine the evaluation phrase set of each label keyword according to the difference value of the tendency coefficient, or to determine the update optimization coefficient of each label keyword according to the evaluation tendency coefficient of the effective matching phrase and the evaluation key coefficient; Determine the text reference status of each relevant analysis text according to the text overlap parameter and the evaluation reference index; If a text overlap parameter of a related analysis text is greater than a preset text overlap parameter or an evaluation reference index is less than or equal to a preset evaluation reference index, the related analysis text is determined to be in a first preset text reference state; If the text overlap parameter of a related analysis text is less than or equal to the preset text overlap parameter and the evaluation reference index is greater than the preset evaluation reference index, the related analysis text is determined to be in the second preset text reference state; Determine effective analysis strategies based on the textual reference status of each relevant analysis text; If a related analysis text is in the first preset text reference state, performing phrase reference analysis on the text analysis set; If a related analysis text is in the second preset text reference state, an effective evaluation analysis is performed on the related evaluation keywords.

2. The adaptive commodity classification method based on AI recognition of a large model according to claim 1 is characterized in that: The division method of the text analysis set is determined according to the text division coefficient; If the text division coefficient of a related analysis text is greater than the preset text division coefficient, a text analysis set is determined according to the evaluation relevance; If the text division coefficient of a related analysis text is less than or equal to the preset text division coefficient, determine the text analysis set according to the evaluation reference tendency coefficient and the reference related evaluation coefficient; The text division coefficient is determined according to the evaluation reference index and the relevant evaluation coefficient.

3. The adaptive commodity classification method based on AI recognition of large models according to claim 2 is characterized in that: For a single text analysis set, the phrase reference analysis process includes: Determine effective evaluation phrases according to text correlation coefficient and related evaluation coefficient; Determine the setting method of the evaluation key coefficient of the effective evaluation phrase according to the division method of the text analysis set; If the text analysis set is determined according to the evaluation relevance, the evaluation key coefficient for each valid evaluation phrase of the text analysis set is set according to the reference coincidence coefficient, and the evaluation key coefficient is negatively correlated with the reference coincidence coefficient; If the text analysis set is determined based on the reference tendency coefficient and the reference correlation coefficient, the evaluation key coefficient for each valid evaluation phrase of the text analysis set is set based on the set abnormal tendency parameter and the set reference correlation coefficient, and the evaluation key coefficient is positively correlated with the set evaluation index.

4. The adaptive commodity classification method based on AI recognition of a large model according to claim 3 is characterized in that: Extracting relevant evaluation phrases from each relevant analysis text in the second preset text reference state, and determining the evaluation validity coefficient of the evaluation phrase according to the text correlation coefficient and the proportion of key users; The evaluation phrases whose evaluation effectiveness coefficient is greater than the preset evaluation effectiveness coefficient are recorded as effective evaluation phrases, and effective evaluation analysis is performed on the evaluation key coefficient of each effective evaluation phrase.

5. The adaptive commodity classification method based on AI recognition of a large model according to claim 4 is characterized in that: The effective evaluation analysis process for a single effective evaluation phrase includes: Detecting the user reference coefficient of the relevant analysis text of the effective evaluation phrase, wherein the user reference coefficient is determined according to the user evaluation quality parameter and the historical category overlap parameter; Determine the evaluation key coefficient of the effective evaluation phrase according to the text relevance and the user reference coefficient; The evaluation key coefficient is positively correlated with the text relevance and the user reference coefficient respectively.

6. The adaptive commodity classification method based on AI recognition of a large model according to claim 5 is characterized in that: When the evaluation extraction is completed, the effective matching phrases of each tag keyword are determined according to the evaluation related frequency and the update overlap ratio, and the tag update strategy is determined according to the effective key parameters and the reference key coefficients; If the effective key parameter is greater than the preset effective key parameter or the reference key coefficient is greater than the preset reference key coefficient, the evaluation phrase set of each tag keyword is determined according to the difference value of the tendency coefficient; If the effective key parameter is less than or equal to the preset effective key parameter and the reference key coefficient is less than or equal to the preset reference key coefficient, then the update optimization coefficient of each tag keyword is determined according to the evaluation tendency coefficient of the effective matching phrase and the evaluation key coefficient; The evaluation extraction completion condition is that all relevant analysis texts of the target category products have completed effective key analysis.

7. The adaptive commodity classification method based on AI recognition of a large model according to claim 6 is characterized in that: For a single tag keyword, determine the update optimization coefficient of each tag keyword based on the tendency conflict coefficient of each evaluation phrase set and the set reference key coefficient; The update optimization coefficient is positively correlated with the tendency conflict coefficient and the set reference key coefficient respectively; The difference value of the tendency coefficient of any evaluation phrase set is smaller than the preset difference value of the tendency coefficient.

8. The adaptive commodity classification method based on AI recognition of a large model according to claim 7 is characterized in that: When determining the update optimization coefficient of each tag keyword according to the evaluation tendency coefficient and the evaluation key coefficient of the effective matching phrase, the update optimization coefficient is determined according to the reference tendency parameter and the relevant key coefficient; The update optimization coefficient is positively correlated with the relevant matching parameter and the relevant key coefficient respectively; The early warning update label is determined according to the update optimization coefficient of each label keyword, and the label keyword whose update optimization coefficient is greater than the preset update optimization coefficient is recorded as the early warning update label.

Images