Method and system for displaying idle items
By extracting the similarity calculation of user-demand keywords and idle item labels, combining dictionary and correlation, optimizing the search path, the problem of blocked information circulation and low matching accuracy in the existing technology is solved, and efficient idle item transactions are achieved. Improves user experience and transaction success rate.
Patent Information
- Application Number
- CN202411302711.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-19
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-19
AI Technical Summary
Due to the improper information circulation and low matching accuracy of the existing idle item trading model, it is difficult for users to quickly find items that suit their wishes in massive information, which reduces transaction efficiency and accuracy, and has high transaction costs.
By obtaining user demand information, extracting keywords and performing similarity calculations with idle item labels, we automatically display items with high matching degrees, combining dictionary and correlation calculations, optimize search paths, reduce manual screening, improve recommendation accuracy, and pass price Setting and market price evaluation to optimize the trading process.
It achieves efficient matching of user needs and idle items, improves transaction efficiency and user satisfaction, reduces transaction costs, and enhances the ease of use of the platform and user loyalty.
Smart Images

Figure CN118822694B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of idle item transaction management, and in particular to an idle item display method and system. Background Art
[0002] With the improvement of social and economic levels and the profound change in consumption concepts, the number of idle items has shown a significant growth trend. How to efficiently manage and make full use of these resources has become an important issue that society needs to solve urgently. The traditional idle item trading model, whether it is a physical store, a second-hand market or interpersonal word-of-mouth communication, is limited by the closedness of information flow, low matching accuracy and high transaction costs, and it is difficult to meet the rapid transaction needs of modern society.
[0003] Entering the digital age, the widespread coverage of the Internet has opened up an unprecedented new world for the trading of idle items. Although many online second-hand trading platforms have emerged on the market, they allow users to obtain item information through keyword searches, but they do not push information in a certain order according to user needs, causing users to spend a lot of time screening the most suitable idle items in a large amount of information, reducing the efficiency and accuracy of finding the desired items, and thus restricting the overall efficiency of transactions. Summary of the invention
[0004] In order to improve the matching degree between idle item information and user demand information and optimize user experience, the present application provides an idle item display method and system.
[0005] In the first aspect, the present application provides a method for displaying idle items, which adopts the following technical solution:
[0006] A method for displaying idle items comprises the following steps:
[0007] Obtaining information: obtaining information of idle items to be traded, adding idle item tags according to the idle item information; obtaining demand information of the first user;
[0008] Building a database: Building a database, wherein the database stores information on idle items to be traded and idle item tags;
[0009] Data processing: including keyword extraction, first calculation and first judgment;
[0010] Extract keywords: extract keywords based on the demand information of the first user and record them as the first keywords;
[0011] First calculation: calculating the similarity between the first keyword and the idle item label, recorded as the first data;
[0012] First judgment: judging whether the first data is greater than a preset similarity threshold, if so, executing the display step; if not, executing the feedback step;
[0013] Display: Push the idle items corresponding to the idle item tags to the first user in descending order of similarity;
[0014] Feedback: There is no such idle item.
[0015] By adopting the above technical solution, the present application can quickly narrow the search scope and reduce the time the first user spends browsing irrelevant idle items by extracting keywords from user needs and calculating similarity with idle item tags in the database. Based on similarity judgment, idle items that best match user needs can be displayed first, which improves the accuracy of recommendations and increases the possibility that the first user will find satisfactory items. The automated processing flow reduces the tediousness of manual screening, allowing the first user to quickly obtain the required idle items, enhancing the platform's ease of use and user satisfaction. Accurate recommendations can stimulate the first user's desire to buy, thereby increasing the chances of a successful transaction.
[0016] Optionally, after the step of performing data processing and before the step of performing feedback, the method further includes:
[0017] Setting dictionary: Setting a dictionary, wherein the dictionary stores the first keyword and the second keyword;
[0018] Second calculation: calculating the similarity between the first keyword and the second keyword in the dictionary to obtain second data;
[0019] Second judgment: judge whether the second data is greater than a preset similarity threshold. If so, use the second keyword corresponding to the maximum similarity as the new first keyword and execute the first calculation step; if not, execute the feedback step.
[0020] By adopting the above technical solution, by introducing a dictionary and calculating the similarity between the first keyword and the second keyword in the dictionary (i.e., the keyword remaining after removing the first keyword from the dictionary), a second keyword that is closer to or more extensive than the user's needs can be identified. If a second keyword with a higher similarity is found, the similarity is recalculated using this second keyword, which helps to more accurately match the real needs of the first user and improve the accuracy of the recommendation. Since the intention of the first user is understood more accurately, idle items that are more in line with the expectations of the first user can be pushed, thereby enhancing the satisfaction and trust of the first user. When using the search function, the first user may find it difficult to directly understand his or her real needs due to unclear expressions, spelling errors, or the use of synonyms. By introducing a dictionary and similarity calculation, these situations can be handled more flexibly, improving the robustness of the system.
[0021] Optionally, after performing the second judgment step and before performing the feedback step, the method further includes:
[0022] Third calculation: calculating the correlation between the first keyword and the second keyword in the dictionary to obtain third data;
[0023] The third judgment: judge whether the third data is greater than the preset correlation threshold. If so, use the second keyword corresponding to the maximum correlation as the new first keyword and execute the first calculation step. If not, execute the feedback step.
[0024] By adopting the above technical solution, by calculating the correlation between the first keyword and the second keyword in the dictionary, the present application can more deeply understand the semantic relationship behind the user query. This correlation is not only based on the surface similarity, but also includes many considerations such as domain knowledge and idioms, so as to more accurately grasp the true intention of the first user. When it is found that there is a high correlation between a second keyword and the first keyword, the present application selects the second keyword with the highest correlation as the new query starting point, and can find content that is more closely related to the user's original needs. This recommendation method based on correlation helps to improve the relevance and usefulness of the recommendation results. Since the recommendation results are closer to the actual needs of the first user, the first user can feel higher satisfaction and convenience, and this positive user experience helps to improve the loyalty of the first user. By continuously iterating the calculation of similarity and correlation, and adjusting the query keywords according to the results, the real needs of the first user can be gradually approached, thereby optimizing the search path and reducing invalid searches and duplication of work.
[0025] Optionally, after executing the step of extracting keywords and before executing the step of first calculating, the method further includes:
[0026] Data collection: collecting unlabeled dialect data, performing word segmentation processing on the dialect data, and obtaining fourth data;
[0027] Adding labels: adding labels to a portion of the fourth data and recording them as fifth data;
[0028] First modeling: Building a CNN model based on deep learning algorithm;
[0029] Pre-training: Use the fourth data to pre-train the CNN model;
[0030] Fine-tuning: Use the fifth data to fine-tune the parameters of the trained CNN model;
[0031] Update: Input the first keyword into the fine-tuned CNN model, output the third keyword, and use the third keyword as the new first keyword.
[0032] By adopting the above technical solution, by collecting unlabeled dialect data and performing word segmentation processing, the present application can process and understand multiple dialects, thereby expanding the scope of application and user groups, and is more suitable for users with strong regional characteristics, which can improve user experience and satisfaction. By adding labels to a portion of the data (i.e., the fifth data) and using these labeled data to fine-tune the CNN model based on the deep learning algorithm, the accuracy and generalization ability of the CNN model can be significantly improved. This supervised learning method helps the CNN model to better capture the inherent laws and characteristics of the data. The first keyword is input into the fine-tuned CNN model, and the third keyword is output as the new first keyword. This process actually converts the original first keyword (dialect) into the third keyword (Mandarin). Since the CNN model can capture complex semantics and contextual information, the output third keyword is often more accurate and in line with the first user's intention.
[0033] Optionally, after the step of performing the updating and before the step of performing the first calculation, the method further includes:
[0034] Second modeling: building n-gram model;
[0035] Evaluation: Input the third keyword output by the CNN model into the n-gram model to evaluate the rationality of the third keyword;
[0036] Word selection: Select the third keyword corresponding to the highest probability as the new first keyword.
[0037] By adopting the above technical solution, the rationality of the third keyword output by the CNN model can be further verified by introducing the n-gram model. The n-gram model is based on a statistical language model and can capture the co-occurrence relationship between words, thereby evaluating the rationality and naturalness of the third keyword in a specific context. In the evaluation process, selecting the third keyword corresponding to the maximum probability as the new first keyword helps to make the final selected third keyword both consistent with the first user's intention and highly rational and natural. This dual screening mechanism helps to improve the overall quality of the third keyword. Since the n-gram model can handle a variety of language phenomena and complex contexts, combining it with the CNN model can enhance the robustness of the system. By improving the accuracy and rationality of the third keyword, the first user's intention can be understood more accurately, and content that better meets user needs can be pushed, which helps to improve the first user experience and satisfaction, and enhance the trust and dependence of the first user.
[0038] Optionally, after performing the step of adding labels and before performing the step of first modeling, the method further includes:
[0039] Back translation: fix the label of the fifth data, back translate the fifth data, and record the back translated fifth data as the seventh data;
[0040] Expansion: Obtain a synonym for the seventh data, and use the synonym and the seventh data as new fifth data.
[0041] By adopting the above technical solution, by back-translating the fifth data and obtaining its synonyms, the data diversity used for model training can be significantly increased. This diversity is not only reflected in vocabulary and expressions, but may also include context and semantic diversity. This increase in diversity helps the CNN model to better generalize to unseen data. Because the training data is more diverse and rich, the CNN model can learn more features and patterns during the training process. This helps to enhance the robustness of the CNN model, so that it can maintain stable performance in the face of various complex situations. Using more diverse data for training can often improve the accuracy and generalization ability of the CNN model. These adjustments can optimize the performance of the CNN model and make it perform better in practical applications. At the same time, because the training data is richer, the CNN model can converge to the optimal solution faster, thereby improving training efficiency.
[0042] Optionally, after executing the first judgment step and before executing the display step, the method further includes:
[0043] Setting price: The second user sets the transaction amount and transaction amount range of the idle items to be traded, and the transaction amount set by the second user is recorded as the first transaction amount.
[0044] By adopting the above technical solution, by allowing the second user to clearly set the transaction amount (first transaction amount) and transaction amount range of the idle items to be traded, the transparency of the transaction can be significantly improved, which enables potential first users to clearly understand the transaction amount of the items, which helps to make more informed purchase decisions. The clarity of price setting helps to enhance the trust between the first user and the second user. The first user can be sure that the second user has made a reasonable valuation of the item, and the second user can also show his sincerity and credibility through a clear price. In the transaction process, the clear setting of the price can reduce the unnecessary time and energy consumption caused by price negotiations. The first user can quickly determine whether the item is in line with his budget, or feedback a price that meets his expectations based on the first transaction amount set by the second user. By comparing the price feedback by the first user with the set price range, it can be determined whether the first user can successfully purchase the idle item, thereby speeding up the transaction process. Setting a price range instead of a single price can provide a certain bargaining space for the first user, while also protecting the second user from the risk of malicious price reduction. This flexible pricing mechanism helps to maintain market fairness and promote the smooth progress of transactions.
[0045] Optionally, after executing the step of setting the price and before executing the step of displaying, the method further includes:
[0046] Obtaining records: searching for the purchase record of the first user, obtaining the transaction amount and the transaction item, recording the transaction amount of the first user as the second transaction amount, and recording the transaction item as the first transaction item;
[0047] Positioning: Using web crawler technology, locate the idle items of the same type as the first transaction item and record them as the first idle items;
[0048] Extracting transaction amounts: extracting transaction amounts of all first idle items, recording them as third transaction amounts, and integrating the third transaction amounts into the first data set;
[0049] First sorting: sorting the third transaction amount in descending order;
[0050] Fourth calculation: Calculate the percentage position of the second transaction amount in the first data set, recorded as the first percentage position, and the calculation model is as follows:
[0051] ;
[0052] Among them, A is the first percentage position; B is the position of the second transaction amount in the first data set; C is the number of third transaction amounts in the same position as the second transaction amount; and D is the total number of third transaction amounts in the first data set.
[0053] By adopting the above technical solution, by obtaining the purchase record of the first user, the consumption habits and purchasing power level of the first user can be understood, and through the first sorting and fourth calculation steps, the percentage position of the second transaction amount (that is, the purchasing power of the first user or the market price level of similar items) in the first data set (the transaction amount of all first idle items) can be calculated, and it is known that the transaction amount of the first user ranks among the total transaction amounts of this type of products.
[0054] Optionally, after executing the fourth calculation step and before executing the display step, the method further includes:
[0055] Obtaining sales prices: Obtaining the transaction amounts of all idle items to be traded, recording them as the fourth transaction amounts, and integrating all the fourth transaction amounts into the second data set;
[0056] Second sorting: sort the fourth transaction amount in descending order;
[0057] Fifth calculation: Calculate the percentage position of the fourth transaction amount in the second data set, recorded as the second percentage position, and the calculation model is as follows:
[0058] ;
[0059] Wherein, E is the second percentage position; F is the position of the first transaction amount in the second data set; G is the number of fourth transaction amounts in the same position as the first transaction amount; H is the total number of fourth transaction amounts in the second data set;
[0060] Fourth judgment: judge whether the difference between the first percentage position and the second percentage position is less than a preset threshold value, if so, execute the display step; if not, execute the feedback step.
[0061] By adopting the above technical solution, by obtaining and sorting the transaction amounts (fourth transaction amounts) of all idle items to be traded, the price distribution of the current market can be more comprehensively understood. Combined with the position (second percentage position) of the first transaction amount (transaction amount set by the second user) in the second data set, the price level of the idle item in the market can be more accurately displayed. By calculating the difference between the first percentage position (based on the purchase record and historical transaction data of the first user) and the second percentage position (based on the price data of all idle items to be traded in the current market), the price set by the second user can be evaluated for rationality. This evaluation helps to ensure that the price set by the second user is in line with the actual market, neither too high nor too low. If the difference between the first percentage position and the second percentage position is less than the preset threshold, it means that the price set by the second user is relatively close to the market price. At this time, directly executing the display step can save time and improve transaction efficiency. If the difference is greater than the preset threshold, the feedback step is executed, thereby reducing the waste of communication time due to unreasonable prices. Secondly, when the first user receives the first transaction amount set by the second user, he is motivated to feedback to the second user that the price setting is too high, and will feedback a psychologically expected quotation to the second user. By calculating the first percentage position, it can be known that the percentage position of the first user's psychologically expected quotation in this type of idle items is approximately known. Based on the percentage position, the first user's quotation for the idle item is converted to obtain the first user's quotation for the idle item. By comparing the quotation with the transaction amount interval set in advance by the second user, if the difference between the first percentage position and the second percentage position is less than the preset threshold, it means that there is a possibility of a successful transaction between the first user and the second user. At this time, the display step is directly executed, which can save time and improve transaction efficiency. If the difference is greater than the preset threshold, the feedback step is executed (that is, the idle item is skipped and not displayed to the first user). The addition of this series of steps makes the transaction process more intelligent and humane. This application can not only help the second user set a reasonable price, but also give timely feedback when the price is unreasonable, help the second user optimize the transaction strategy, save transaction time, and thus improve the user experience.
[0062] In the second aspect, the present application provides an idle item display system, which adopts the following technical solution:
[0063] An idle item display system, the system comprising:
[0064] An information acquisition module is used to acquire information of idle items to be traded, add idle item tags according to the idle item information, and acquire demand information of the first user;
[0065] A database building module, used to build a database, wherein the database stores information on idle items to be traded and idle item tags;
[0066] A data processing module, comprising a keyword extraction unit, a first calculation unit and a first judgment unit;
[0067] A keyword extraction unit, used for extracting a keyword based on the demand information of the first user, and recording the keyword as the first keyword;
[0068] A first calculation unit, used to calculate the similarity between the first keyword and the idle item tag, recorded as first data;
[0069] A first judging unit, used to judge whether the first data is greater than a preset similarity threshold;
[0070] A display module, used to push the idle items corresponding to the idle item tags to the first user in descending order of similarity;
[0071] Feedback module, used to report that there are no such idle items.
[0072] By adopting the above technical solution, by accurately extracting user demand keywords and performing efficient similarity calculations with idle item tags, accurate matching and personalized display of user needs are achieved, which not only improves the user experience and enables the first user to find the desired items faster, but also promotes the effective use of idle resources.
[0073] In summary, the present application includes at least one of the following beneficial technical effects:
[0074] 1. This application can quickly narrow the search scope and reduce the time the first user spends browsing irrelevant idle items by extracting keywords from user needs and calculating similarity with idle item tags in the database. Based on similarity judgment, idle items that best match user needs can be displayed first, which improves the accuracy of recommendations and increases the possibility that the first user will find satisfactory items. The automated processing flow reduces the tediousness of manual screening, allowing the first user to quickly obtain the required idle items, enhancing the platform's ease of use and user satisfaction. Accurate recommendations can stimulate the first user's desire to buy, thereby increasing the chances of a successful transaction.
[0075] 2. By obtaining and sorting the transaction amounts (fourth transaction amounts) of all idle items to be traded, the price distribution of the current market can be more comprehensively understood. Combined with the position (second percentage position) of the first transaction amount (transaction amount set by the second user) in the second data set, the price level of the idle item in the market can be more accurately displayed. By calculating the difference between the first percentage position (based on the purchase record and historical transaction data of the first user) and the second percentage position (based on the price data of all idle items to be traded in the current market), the price set by the second user can be evaluated for rationality. This evaluation helps to ensure that the price set by the second user is in line with market reality, neither too high nor too low. If the difference between the first percentage position and the second percentage position is less than the preset threshold, it means that the price set by the second user is close to the market price. At this time, directly executing the display step can save time and improve transaction efficiency. If the difference is greater than the preset threshold, the feedback step is executed to reduce the waste of communication time due to unreasonable prices. Secondly, when the first user receives the first transaction amount set by the second user, he is motivated to feedback to the second user that the price setting is too high, and will feedback a psychologically expected quotation to the second user. By calculating the first percentage position, it can be known that the percentage position of the first user's psychologically expected quotation in this type of idle items is approximately known. Based on the percentage position, the first user's quotation for the idle item is converted to obtain the first user's quotation for the idle item. By comparing the quotation with the transaction amount interval set in advance by the second user, if the difference between the first percentage position and the second percentage position is less than the preset threshold, it means that there is a possibility of a successful transaction between the first user and the second user. At this time, the display step is directly executed, which can save time and improve transaction efficiency. If the difference is greater than the preset threshold, the feedback step is executed (that is, the idle item is skipped and not displayed to the first user). The addition of this series of steps makes the transaction process more intelligent and humane. This application can not only help the second user set a reasonable price, but also give timely feedback when the price is unreasonable, help the second user optimize the transaction strategy, save transaction time, and thus improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] Figure 1 is a flow chart of Example 1 of the present application;
[0077] Figure 2 This is a flowchart from the first judgment of S33 to the feedback of S5 in Example 1 of the present application;
[0078] Figure 3 This is a flowchart from S31 extracting keywords to S32 first calculation in Example 2 of the present application;
[0079] Figure 4 This is the flowchart shown from the first judgment of S33 to S4 in Example 3 of the present application. DETAILED DESCRIPTION
[0080] The following combination Figures 1 to 4 This application is described in further detail.
[0081] Embodiment 1: This embodiment discloses a method for displaying idle items, referring to Figure 1 The method includes: S1 obtaining information, S2 building a database, S3 data processing, S4 display and S5 feedback. In this embodiment, the idle item information to be traded is first obtained, and the idle item label is added to it, and the demand information of the first user is obtained. Then, a database is established to store the idle item information and the idle item label, and then the similarity between the two is calculated. According to the similarity result, it is determined whether to display the idle item to the first user. The specific steps of this embodiment are as follows:
[0082] S1 obtains information, obtains information about idle items to be traded, adds idle item tags according to the idle item information; obtains demand information of the first user. The idle item information refers to: collecting various information about idle items from the second user, such as the name, description, brand, model, newness, price, picture, etc. of the items. After obtaining the idle item information, this step will automatically add tags to these items according to certain rules or algorithms. These tags are usually short descriptions of the characteristics, attributes or categories of the items, such as "second-hand mobile phones", "brand bags", "children's toy cars", etc.
[0083] This step also obtains the first user's demand information based on the keywords, descriptions, or some screening conditions selected by the first user, such as price range, brand preference, item status, etc.
[0084] S2: Building a database, in which the idle item information and idle item tags to be traded are stored.
[0085] S3 data processing includes S31 keyword extraction, S32 first calculation and S33 first judgment.
[0086] S31 extracts keywords based on the demand information of the first user, adopts natural language processing (NLP) technology or a simple string matching algorithm, extracts keywords, and records the extracted keywords as first keywords.
[0087] For example, the method of extracting keywords using a string matching algorithm is as follows:
[0088] The first user's demand information is: looking for a waterproof backpack suitable for outdoor hiking, with a capacity of 30-40 liters and preferably black or dark blue color.
[0089] The first user's demand information is segmented into words: looking for a waterproof backpack suitable for outdoor hiking, the capacity is preferably between 30-40 liters, and the color is preferably black or dark blue.
[0090] Remove stop words: search, outdoor, hiking, waterproof, backpack, capacity, 30-40 liters, color, black, dark blue.
[0091] Based on the string matching algorithm, the following keywords are directly extracted: outdoor, hiking, waterproof, backpack, capacity 30-40 liters, black, dark blue, and the above keywords are recorded as the first keywords.
[0092] For another example, the method of extracting keywords using natural language processing (NLP) technology is as follows:
[0093] The first user's demand information is segmented into words: looking for a waterproof backpack suitable for outdoor hiking, the capacity is preferably between 30-40 liters, and the color is preferably black or dark blue.
[0094] Identify the part of speech of each word in the sentence (such as noun, verb, adjective, etc.): find / verb, a / quantifier, suitable / verb, outdoor / noun, hiking / verb, of / particle, waterproof / adjective, backpack / noun, / punctuation, capacity / noun, best / adverb, in / preposition, 30-40 liters / quantifier, between / preposition, / punctuation, color / noun, best / adverb, is / predicate, black / noun, or / conjunction, dark blue / noun.
[0095] Extract keywords based on part of speech and context. Nouns and adjectives play an important role in describing items, attributes, or concepts, so they are more likely to become keywords. Verbs, although important, are not as directly relevant as nouns and adjectives when extracting keywords for product attributes.
[0096] In the example above, the noun "backpack" is obviously a keyword because it is the item the user is looking for. The adjective "waterproof" and the nouns "capacity" and "color" are also keywords because they describe the attributes of the backpack. Although "outdoor hiking" is a noun phrase, it describes more about the use scenario of the backpack than the attributes of the backpack itself, but it is still an important context clue.
[0097] Context can help understand the specific meaning and importance of words in a sentence. In “Looking for a waterproof backpack suitable for outdoor hiking”, “waterproof” and “backpack” are closely connected, indicating that the first user hopes that the backpack has a waterproof function.
[0098] “The capacity should preferably be between 30-40 liters” further refines the first user's demand for backpack capacity.
[0099] “The color is preferably black or dark blue” indicates the color preferred by the first user.
[0100] Combining part of speech and context information, the following keywords are extracted:
[0101] Backpack (noun, directly refers to the item the first user wants to find);
[0102] Waterproof (an adjective describing an important attribute of a backpack);
[0103] capacity (noun, related to the capacity property of a backpack);
[0104] 30-40 liters (quantity phrase, specifically describing the range of capacity);
[0105] color (noun, related to the color attribute of the backpack);
[0106] black (noun, one of the colors preferred by the first user);
[0107] Dark blue (noun, another color preferred by the first user).
[0108] S32 first calculation, calculating the similarity between the first keyword and the idle item label, recorded as first data. The similarity calculation method includes text similarity calculation based on cosine similarity, string similarity calculation based on edit distance, etc. The calculated similarity value (first data) reflects the matching degree between the user demand and the idle item.
[0109] For example: the first keyword is a keyword extracted using natural language processing (NLP) technology. Suppose the label of an idle item is: "waterproof backpack, 40 liters, black".
[0110] The Word2Vec model is used to convert the first keyword and idle item labels into word vectors, as shown in Table 1.
[0111] Table 1 Schematic table of word vectors corresponding to the first keyword or idle item label
[0112]
[0113] The text similarity calculation method of cosine similarity is as follows:
[0114] The similarity between "waterproof backpack" and "backpack" is:
[0115] ;
[0116] Similarly, we can calculate:
[0117] The similarity between “waterproof backpack” and “waterproof” is: The similarity between "waterproof backpack" and "capacity" is: The similarity between "waterproof backpack" and "30-40 liters" is: The similarity between “waterproof backpack” and “color” is: The similarity between “waterproof backpack” and “black” is: The similarity between "waterproof backpack" and "dark blue" is: .
[0118] The similarity between "40 liters" and "backpack" is: The similarity between "40 liters" and "waterproof" is: The similarity between "40 liters" and "capacity" is: The similarity between "40 liters" and "30-40 liters" is: The similarity between "40 liters" and "color" is: The similarity between "40 liters" and "black" is: The similarity between "40 liters" and "dark blue" is: .
[0119] The similarity between "black" and "backpack" is: The similarity between "black" and "waterproof" is: The similarity between “black” and “capacity” is: The similarity between "black" and "30-40 liters" is: . The similarity between "black" and "color" is: . The similarity between "black" and "black" is: . The similarity between "black" and "dark blue" is: .
[0120] Add up the maximum similarity between each first keyword and the idle item tag to get the total similarity value:
[0121] The total similarity = 0.9985+0.97+0.76+0.99+0.99+1.00+0.99=6.6985.
[0122] This total similarity is the first data.
[0123] The string similarity calculation method based on edit distance is as follows:
[0124] Converting "Waterproof Backpack" to "Backpack" requires deleting the word "waterproof", which means two deletion operations are required. Therefore, the edit distance between "Waterproof Backpack" and "Backpack" is 2, and the similarity is calculated using the following formula:
[0125] ;
[0126] in,
[0127] , , ,
[0128] So the similarity is:
[0129] .
[0130] Similarly, we can calculate:
[0131] The similarity between "waterproof backpack" and "waterproof" is about 0.67. The similarity between "waterproof backpack" and "capacity" is about 0.17. The similarity between "waterproof backpack" and "30-40 liters" is about 0.40. The similarity between "waterproof backpack" and "color" is about 0.33. The similarity between "waterproof backpack" and "black" is about 0.33. The similarity between "waterproof backpack" and "dark blue" is 0.44.
[0132] The similarity between "40 liters" and "backpack" is 0.00. The similarity between "40 liters" and "waterproof" is 0.00. The similarity between "40 liters" and "capacity" is about 0.25. The similarity between "40 liters" and "30-40 liters" is about 0.67. The similarity between "40 liters" and "color" is 0.00. The similarity between "40 liters" and "black" is 0.00. The similarity between "40 liters" and "dark blue" is 0.00.
[0133] The similarity between "black" and "backpack" is about 0.33. The similarity between "black" and "waterproof" is about 0.67. The similarity between "black" and "capacity" is about 0.67. The similarity between "black" and "30-40 liters" is about 0.25. The similarity between "black" and "color" is about 0.83. The similarity between "black" and "black" is 1. The similarity between "black" and "dark blue" is about 0.71.
[0134] The total similarity = 0.5+0.67+0.67+0.67+0.83+1+0.71=5.05.
[0135] This total similarity is the first data.
[0136] S33 is the first judgment, judging whether the first data is greater than a preset similarity threshold. If so, it means that the idle item has a high matching degree with the demand information of the first user, and S4 is executed for display; if not, it means that no suitable idle item is found, and S5 is executed for feedback.
[0137] S4 shows that the idle items corresponding to the idle item tags are pushed to the first user in order of similarity from large to small. This sorting method helps the first user find the items that best meet his needs more quickly.
[0138] S5 feedback, feedback that there is no such idle item, through feedback helps the first user to obtain the search results of the system and prompts the first user to adjust his own needs.
[0139] This embodiment first collects information on idle items to be traded and adds tags, and obtains user needs at the same time; then, this information is stored in a database for retrieval; by extracting keywords of the first user's needs and calculating the similarity between these first keywords and the idle item tags, matching idle items are screened; if items with a similarity higher than a preset threshold are found, they are displayed to the first user in descending order of similarity; if no matching items are found, information that there is no such idle item is fed back to the first user. This embodiment can improve the matching degree between idle item information and user needs and optimize user experience.
[0140] like Figure 2 As shown, in other embodiments, after performing S3 data processing and before performing S5 feedback, it also includes:
[0141] S34 sets a dictionary, wherein the dictionary stores a first keyword and a second keyword. The first keyword refers to the first keyword extracted in S31, and the second keyword refers to all keywords remaining in the dictionary after removing the first keyword. The second keyword comes from the first keyword extracted and stored in the historical transaction process.
[0142] S35 is a second calculation, which is to calculate the similarity between the first keyword in the keywords extracted in S31 and the second keyword in the dictionary to obtain second data.
[0143] S36 is the second judgment, judging whether the second data is greater than the preset similarity threshold. If so, it means that the keyword is highly relevant to the needs of the first user. The second keyword corresponding to the maximum similarity is used as the new first keyword, and the first calculation of S32 is executed; if not, the third calculation of S37 is executed.
[0144] The similarity threshold preset in this step is the same as the similarity threshold preset in the first judgment in S33.
[0145] S37 third calculation: if the second keyword with sufficiently high similarity is not found in S36 second judgment, the correlation between the first keyword and the second keyword in the dictionary (third data) is calculated instead. The calculation of the correlation takes into account more contextual information, historical search data or user behavior patterns, aiming to mine potential, less direct but still relevant search options.
[0146] For example, the first keyword is a keyword extracted using natural language processing (NLP) technology, and the second keyword is assumed to be: "shoulder bag, black and white".
[0147] The Word2Vec model is used to convert the second keyword into a word vector, as shown in Table 2.
[0148] Table 2 Schematic diagram of word vectors corresponding to the second keyword
[0149]
[0150] The correlation is calculated using the Pearson correlation coefficient, and the calculation model is as follows:
[0151] .
[0152] The correlation between "backpack" and "shoulder bag" is calculated as follows:
[0153] ;
[0154] ;
[0155] ;
[0156] ;
[0157] ;
[0158] ;
[0159] The correlation .
[0160] Similarly, the relevance between each first keyword and the second keyword is calculated, and the maximum relevance with the second keyword "shoulder bag" and "black and white" is obtained and added up to obtain the total relevance, which is recorded as the third data.
[0161] S38 is the third judgment, judging whether the third data is greater than the preset correlation threshold. If so, the second keyword corresponding to the maximum correlation is used as the new first keyword, and the first calculation of S32 is executed. If not, S5 is executed for feedback.
[0162] For example: the first user's demand information is: "black marble desktop", and its first keywords are "black", "marble" and "desktop". After calculating the similarity, it is found that there is no matching second keyword, so the correlation between the second keyword and the first keyword is calculated. After calculation, it is found that the correlation between "black", "stone" and "table" exceeds the preset correlation threshold, then it is used as the new first keyword and S32 first calculation is re-executed.
[0163] This embodiment constructs a keyword dictionary and iteratively optimizes the search keywords by calculating similarity and relevance to find idle items that best match or are highly relevant to the original needs of the first user. If a keyword that meets the threshold is successfully found, the first calculation in S32 is performed again; if not found, S5 feedback is executed to display no matching results to the user or provide search suggestions, thereby improving search accuracy and user satisfaction.
[0164] Example 2: Figure 3 As shown, the difference from Embodiment 1 is that after executing S31 to extract keywords and before executing S32 to perform the first calculation, it also includes:
[0165] S601 data collection, collecting unlabeled dialect data, performing word segmentation processing on the dialect data, and obtaining fourth data.
[0166] S602 adds a label, adds a label to a portion of the fourth data, the label is the Mandarin corresponding to the fourth data, and records the fourth data and the label together as fifth data.
[0167] S603 back-translation, fixing the label of the fifth data, back-translating the fifth data, and recording the back-translated fifth data as the seventh data. In this step, the labels of the seventh data and the fifth data are the same.
[0168] S604 expansion, obtaining a synonym for the seventh data, and using the synonym and the seventh data as new fifth data. In this step, the synonym is also provided with the same label as the seventh data.
[0169] S605: First modeling, building a CNN model based on a deep learning algorithm. The CNN model includes: an input layer and a hidden layer, and the hidden layer includes: a convolutional layer, a pooling layer, and a fully connected layer.
[0170] S606 pre-training, using the fourth data to pre-train the CNN model, so that the CNN model learns the patterns and features of the dialect data.
[0171] The training process is as follows:
[0172] S6061. Input the fourth data into the CNN model. The fourth data is sent to the input layer of the CNN model.
[0173] S6062. Calculate the loss (such as cross entropy loss for classification tasks) through forward propagation, and adjust the CNN model parameters through the back-propagation algorithm to minimize the loss.
[0174] Convolutional layer: The output of the input layer is convolved with the convolution kernel to generate feature maps. These feature maps are passed through an activation function (such as ReLU) to increase nonlinearity.
[0175] Pooling layer: Performs pooling operations (such as maximum pooling or average pooling) on the feature maps output by the convolutional layer to reduce the spatial dimension of the data and reduce the amount of computation.
[0176] Fully connected layer: After the convolutional layer and the pooling layer, a fully connected layer is set to map the learned features to the output space.
[0177] S6063. Repeat S6061 to S6063 until the CNN model shows good performance on the validation set, that is, the CNN model has learned the patterns and features of the dialect data.
[0178] S607 fine-tuning, using the fifth data to fine-tune the parameters of the trained CNN model, so that the trained CNN model has the ability to convert the dialect into Mandarin.
[0179] In this embodiment, the cross entropy loss function is used as the loss function for CNN model training, and the parameters of the CNN model are updated by the gradient descent method until a minimized loss function is obtained. Specifically, for each parameter, its current value is subtracted from the product of the learning rate and the gradient to obtain a new parameter value.
[0180] Based on the fifth data expanded in S604, the label of the fifth data is recorded as the true label, and the calculation model of the loss function is as follows:
[0181] ;
[0182] Wherein, N is the number of samples in the fifth data, is the true label of sample i on the output word j (the value is 0 or 1, 0 means the true label does not belong to the word, 1 means the true label belongs to the word), It represents the probability that sample i belongs to word j predicted by the CNN model.
[0183] The parameters of the CNN model are updated by the gradient descent method as follows:
[0184] For the parameters of the CNN model Its update rule is:
[0185] .
[0186] in, is the learning rate, set to 0.01; is the gradient, and its calculation model is as follows:
[0187] .
[0188] S608 is an update step, where the first keyword is input into the fine-tuned CNN model, a third keyword is output, and the third keyword is used as a new first keyword. The third keyword mentioned in this step is the Mandarin corresponding to the first keyword.
[0189] S609: Second modeling: establishing an n-gram model.
[0190] S610 evaluation: input the third keyword output by the CNN model into the n-gram model to evaluate the rationality of the third keyword.
[0191] The n-gram model is a tool based on statistical language models that is used to predict the sequence of words in a text. The third keyword output by the CNN model is put into the n-gram model to evaluate its rationality in a given context (or the entire text), which is achieved in the following way:
[0192] Context simulation: Construct a context containing the first few third keywords as the input of the n-gram model.
[0193] Probability evaluation: Use the n-gram model to calculate the probability of the third keyword appearing in a given context. The greater the output probability, the more reasonable the third keyword is.
[0194] The process of using the n-gram model to evaluate the rationality of the third keyword is as follows:
[0195] Using the training corpus, an n-gram model is constructed. The n-gram model learns the co-occurrence frequency between w1 and w2.
[0196] Using Bayes' theorem, calculate the conditional probability P(w2|w1). This probability represents the possibility of w2 appearing under the condition that w1 is known.
[0197] P(w2|w1) = P(w1, w2) / P(w1).
[0198] When the conditional probability P(w2|w1) exceeds the preset threshold, it is considered reasonable, otherwise it is considered unreasonable.
[0199] S611 selects a word, and selects the third keyword corresponding to the maximum probability output by the n-gram model as the new first keyword.
[0200] For example, the dialect term "gaizhui" is predicted by the CNN model as "screwdriver", and this conversion is determined to be reasonable by the n-gram model.
[0201] In this embodiment, after extracting keywords and before performing the first calculation, a dialect data processing and model optimization process is added, including collecting dialect data, word segmentation, label addition, backtranslation, expansion, CNN model construction and pre-training, fine-tuning, keyword update, and evaluating and selecting words through the n-gram model to ensure that the output keywords are both accurate and conform to language norms, providing high-quality input for subsequent data processing.
[0202] Embodiment 3: As Figure 4 shown, the difference between this embodiment and Embodiment 1 is that after performing the first judgment in S33 and before performing S4 display, it further includes:
[0203] S701 Set price: The second user sets the transaction amount and the transaction amount range of the idle item to be traded, and records the transaction amount set by the second user as the first transaction amount. The transaction amount is known to the first user, and the transaction amount range is not known to the first user. The transaction amount range is used to compare with the first user's feedback offer to determine whether the idle item can continue to be displayed to the first user.
[0204] S702 Obtain records, search for the purchase records of the first user, obtain the transaction amount and the transaction item, record the transaction amount of the first user as the second transaction amount, and record the transaction item as the first transaction item.
[0205] S703 Locate, use web crawler technology to locate idle items of the same type as the first transaction item, and record them as the first idle items.
[0206] S704 Extract transaction amount, extract the transaction amounts of all the first idle items, record them as the third transaction amount, and integrate the third transaction amount into the first data set.
[0207] S705 First sorting, sort the third transaction amount in descending order.
[0208] S706 Fourth calculation, calculate the percentage position of the second transaction amount in the first data set, and record it as the first percentage position. The calculation model is as follows:
[0209] ;
[0210] Among them, A is the first percentage position; B is the position of the second transaction amount in the first data set; C is the number of third transaction amounts in the first data set that are in the same position as the second transaction amount; and D is the total number of third transaction amounts in the first data set.
[0211] The first percentage position reflects the relative position of the first user's purchasing power with respect to the prices of similar idle items in the market.
[0212] S707 obtains the sales price, obtains the transaction amount of all the idle items to be traded, records it as the fourth transaction amount, and integrates all the fourth transaction amounts into a second data set.
[0213] S708 second sorting, sorting the fourth transaction amounts in descending order.
[0214] S709 is a fifth calculation, which is to calculate the percentage position of the fourth transaction amount in the second data set, recorded as the second percentage position, and the calculation model is as follows:
[0215] ;
[0216] Among them, E is the second percentage position; F is the position of the first transaction amount in the second data set; G is the number of fourth transaction amounts in the second data set that are in the same position as the first transaction amount; H is the total number of fourth transaction amounts in the second data set.
[0217] The second percentage position reflects the relative position of the price set by the second user with respect to the overall market price.
[0218] S710 is the fourth judgment, judging whether the difference between the first percentage position and the second percentage position is less than a preset threshold value, which represents the acceptable range of the price acceptance difference between the buyer and the seller. If so, it means that the price expectations of the buyer and the seller are relatively close, and S4 is executed for display; if not, S5 is executed for feedback.
[0219] For example, for idle item Q, the second user marked the price as 100, and the price range was set to 90-100. The quotations for all idle items Q to be traded were: 87, 91, 99, 100, 120. The second percentage position calculation formula is as follows:
[0220] .
[0221] After calculation, it is found that the first percentage position of the first user is 30%, that is, it is predicted that the expected price of the first user is lower than the lowest value of the price range by 90%.
[0222] The preset threshold is 10%, and the difference between the first percentage position and the second percentage position is 5%, which means that the first user is likely to purchase the idle item Q posted by the second user, and the idle item Q will be displayed.
[0223] When the difference between the first percentage position and the second percentage position is greater than 10%, that is, when the first percentage position is greater than 35%, it indicates that the probability of a successful transaction between the first user and the second user is relatively low, and the idle item Q posted by the second user is not recommended to the first user.
[0224] When the price expected by the first user is higher than the transaction amount preset by the second user, the idle item will be directly displayed to the first user, and the relevant steps of this embodiment will no longer be executed.
[0225] This embodiment sets the transaction amount, obtains user purchase records, locates similar idle items, extracts and sorts the transaction amounts, calculates the percentage position of the buyer and seller's prices in the market, and determines whether the difference between the two is less than a preset threshold to decide whether to display transaction information or provide feedback, thereby optimizing transaction matching and user experience.
[0226] Embodiment 4: This embodiment discloses an idle item display system, the system comprising:
[0227] The information acquisition module is used to acquire the idle item information to be traded, add idle item tags according to the idle item information, and acquire the demand information of the first user. The tags are divided based on multiple dimensions such as the type, brand, function, and newness of the item, so as to facilitate subsequent data processing and user search.
[0228] The database building module is used to build a database, in which the idle item information and idle item tags to be traded are stored.
[0229] The data processing module includes a keyword extraction unit, a first calculation unit and a first judgment unit.
[0230] The keyword extraction unit is used to extract keywords related to idle items, i.e., first keywords, from the demand information of the first user by using technologies such as natural language processing. These keywords represent the search intention and preference of the user.
[0231] The first calculation unit is responsible for calculating the similarity between the first keyword and each idle item tag in the database. This similarity is calculated based on a certain algorithm (such as cosine similarity, Jaccard similarity, etc.), which reflects the matching degree between user needs and idle items. The calculated result is recorded as the first data, that is, the similarity value.
[0232] The first judgment unit is used to judge whether the first data is greater than a preset similarity threshold. If the similarity value is greater than the threshold, it means that the idle item is highly matched with the user's needs; otherwise, it means that the matching degree is not high.
[0233] The display module is used to push the idle items corresponding to the idle item tags to the first user in the order of similarity from large to small.
[0234] Feedback module, used to report that there are no such idle items.
[0235] This system includes collecting and labeling idle item information, establishing a database, processing user needs to extract keywords and calculate the similarity with item tags, and finally pushing matching idle items to users based on similarity sorting. If no match is found, feedback that there is no such item is provided, thereby providing an efficient and personalized transaction experience.
[0236] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.
Claims
1. A method for displaying idle items, characterized in that: include: Obtaining information: obtaining information of idle items to be traded, adding idle item tags according to the idle item information; obtaining demand information of the first user; Building a database: Building a database, wherein the database stores information on idle items to be traded and idle item tags; Data processing: including keyword extraction, first calculation and first judgment; Extract keywords: extract keywords based on the demand information of the first user and record them as the first keywords; First calculation: calculating the similarity between the first keyword and the idle item label, recorded as the first data; First judgment: judging whether the first data is greater than a preset similarity threshold, if so, executing the step of setting a price; if not, executing the step of providing feedback; Setting price: the second user sets the transaction amount and transaction amount range of the idle items to be traded, and the transaction amount set by the second user is recorded as the first transaction amount; Obtaining records: searching for the purchase record of the first user, obtaining the transaction amount and the transaction item, recording the transaction amount of the first user as the second transaction amount, and recording the transaction item as the first transaction item; Positioning: Using web crawler technology, locate the idle items of the same type as the first transaction item and record them as the first idle items; Extracting transaction amounts: extracting transaction amounts of all first idle items, recording them as third transaction amounts, and integrating the third transaction amounts into the first data set; First sorting: sorting the third transaction amount in descending order; Fourth calculation: Calculate the percentage position of the second transaction amount in the first data set, recorded as the first percentage position, and the calculation model is as follows: ; Wherein, A is the first percentage position; B is the position of the second transaction amount in the first data set; C is the number of third transaction amounts in the same position as the second transaction amount; D is the total number of third transaction amounts in the first data set; Obtaining sales prices: Obtaining the transaction amounts of all idle items to be traded, recording them as the fourth transaction amounts, and integrating all the fourth transaction amounts into the second data set; Second sorting: sort the fourth transaction amount in descending order; Fifth calculation: Calculate the percentage position of the fourth transaction amount in the second data set, recorded as the second percentage position, and the calculation model is as follows: ; Wherein, E is the second percentage position; F is the position of the first transaction amount in the second data set; G is the number of fourth transaction amounts in the same position as the first transaction amount; H is the total number of fourth transaction amounts in the second data set; Fourth judgment: judging whether the difference between the first percentage position and the second percentage position is less than a preset threshold, if so, executing the display step; if not, executing the feedback step; Display: Push the idle items corresponding to the idle item tags to the first user in descending order of similarity; Feedback: There is no such idle item.
2. The idle item display method according to claim 1, characterized in that: After the step of performing data processing and before the step of performing feedback, the method further includes: Setting dictionary: Setting a dictionary, wherein the dictionary stores the first keyword and the second keyword; Second calculation: calculating the similarity between the first keyword and the second keyword in the dictionary to obtain second data; Second judgment: judge whether the second data is greater than a preset similarity threshold. If so, use the second keyword corresponding to the maximum similarity as the new first keyword and execute the first calculation step; if not, execute the feedback step.
3. The idle item display method according to claim 2, characterized in that: After executing the second judgment step and before executing the feedback step, the method further includes: Third calculation: calculating the correlation between the first keyword and the second keyword in the dictionary to obtain third data; The third judgment: judge whether the third data is greater than the preset correlation threshold. If so, use the second keyword corresponding to the maximum correlation as the new first keyword and execute the first calculation step. If not, execute the feedback step.
4. The idle item display method according to any one of claims 1 to 3, characterized in that: After executing the step of extracting keywords and before executing the step of first calculating, the method further includes: Data collection: collecting unlabeled dialect data, performing word segmentation processing on the dialect data, and obtaining fourth data; Adding a label: adding a label to a portion of the fourth data, wherein the label is the Mandarin corresponding to the fourth data, and the fourth data and the label are recorded together as the fifth data; First modeling: Building a CNN model based on deep learning algorithm; Pre-training: Use the fourth data to pre-train the CNN model; Fine-tuning: Use the fifth data to fine-tune the parameters of the trained CNN model; Update: Input the first keyword into the fine-tuned CNN model, output the third keyword, and use the third keyword as the new first keyword.
5. The idle item display method according to claim 4, characterized in that: After the step of performing the updating and before the step of performing the first calculation, the method further includes: Second modeling: building n-gram model; Evaluation: Input the third keyword output by the CNN model into the n-gram model to evaluate the rationality of the third keyword; Word selection: Select the third keyword corresponding to the highest probability as the new first keyword.
6. The idle item display method according to claim 5, characterized in that: After performing the step of adding labels and before performing the step of first modeling, the method further includes: Back translation: fix the label of the fifth data, back translate the fifth data, and record the back translated fifth data as the seventh data; Expansion: Obtain a synonym for the seventh data, and use the synonym and the seventh data as new fifth data.
7. An idle item display system, characterized in that: The system is used to implement the method according to any one of claims 1 to 6, and the system further includes: An information acquisition module is used to acquire information of idle items to be traded, add idle item tags according to the idle item information, and acquire demand information of the first user; A database building module, used to build a database, wherein the database stores information on idle items to be traded and idle item tags; A data processing module, comprising a keyword extraction unit, a first calculation unit and a first judgment unit; A keyword extraction unit, used for extracting a keyword based on the demand information of the first user, and recording the keyword as the first keyword; A first calculation unit, used to calculate the similarity between the first keyword and the idle item tag, recorded as first data; A first judging unit, used to judge whether the first data is greater than a preset similarity threshold; A display module, used to push the idle items corresponding to the idle item tags to the first user in descending order of similarity; Feedback module, used to report that there are no such idle items.
