Product replenishment method and device, electronic equipment and computer readable medium

Abstract

The embodiment of the invention discloses a product replenishment method and device, electronic equipment and a computer readable medium. A specific embodiment of the method comprises the following steps: acquiring a product information set; preprocessing each piece of product information to generate smooth flow volumes and product cleaning names, and obtaining a smooth flow volume set and a product cleaning name set; inputting the product cleaning name set and the target product name into a pre-training model to obtain a product name vector set and a target product name vector; determining similar information of each product name vector and the target product name vector, and generating a future product flow volume; and controlling a printing device corresponding to the target product to print a product title report corresponding to the future product flow volume so as to schedule a replenishment machine device and realize replenishment processing for the future product flow volume. According to the embodiment, timely replenishment processing is carried out according to the predicted product flow volume, and reasonable utilization of storage resources is realized.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and more particularly to a product replenishment method, apparatus, electronic device, and computer-readable medium. Background Art

[0002] The speed of product replacement can impact a company's competitiveness. Rapid product iterations complicate replenishment decisions, leading to inventory backlogs and stockouts, which in turn impact a company's ability to control the market. Rapidly iterating new product turnover forecasts to facilitate replenishment have become a necessity. Forecasting turnover for rapidly iterating new products typically involves using the time series patterns of new product turnover over a given lifecycle to predict full-cycle turnover demand.

[0003] However, when using the above method to predict product turnover, the following technical problems often arise:

[0004] First, new products lack historical turnover data, and methods that rely on time series patterns or label training are ineffective, resulting in deviations in replenishment quantity forecasts, causing inventory imbalances and reduced warehouse scheduling efficiency.

[0005] Second, the data in the product management system is chaotic, which causes the expert system to make rough predictions based on the category dimensions, causing the replenishment quantity to deviate from the actual demand and the warehouse inventory structure to be unbalanced.

[0006] Third, with massive products and complex scenarios, the use of manual data labeling or calculations is complex and computationally intensive, and the cost is high, and cannot meet the real-time scheduling needs of warehouse replenishment.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the Invention

[0008] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0009] Some embodiments of the present disclosure provide a product replenishment method, apparatus, electronic device, and computer-readable medium to solve one or more of the technical problems mentioned in the above background technology section.

[0010] In a first aspect, some embodiments of the present disclosure provide a product replenishment method, comprising: obtaining a product information set, wherein each product information in the product information set includes a product name and a product turnover; preprocessing each product information in the product information set to generate a smoothed turnover and a product cleaning name, thereby obtaining a smoothed turnover set and a product cleaning name set; inputting the product cleaning name set and a target product name into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector, wherein the target product is a product for which future turnover is to be predicted; determining similarity information between each product name vector in the product name vector set and the target product name vector to obtain a similarity information set; generating a future product turnover corresponding to the target product based on the smoothed turnover set and the similarity information set; controlling a printing device corresponding to the target product to print a product title report corresponding to the future product turnover, dynamically adjusting a component position of a product display component corresponding to the target product on a product display page based on the future turnover, and controlling a product storage device to schedule a replenishment machine device to implement replenishment processing for the future product turnover.

[0011] In a second aspect, some embodiments of the present disclosure provide a product replenishment device, comprising: an acquisition unit configured to acquire a product information set, wherein each product information in the above product information set includes a product name and a product turnover; a preprocessing unit configured to preprocess each product information in the above product information set to generate a smooth turnover and a product cleaning name, thereby obtaining a smooth turnover set and a product cleaning name set; a vectorization unit configured to input the above product cleaning name set and the target product name into a pre-trained model for product name vector conversion, thereby obtaining a product name vector set and a target product name vector, wherein the above target product is a product for which future turnover is to be predicted; similar information determination The unit is configured to determine the similarity information between each product name vector in the above-mentioned product name vector set and the above-mentioned target product name vector to obtain a similarity information set; the future turnover prediction unit is configured to generate the future product turnover corresponding to the above-mentioned target product based on the above-mentioned smoothed turnover set and the above-mentioned similarity information set; the printing device unit is configured to control the printing device corresponding to the above-mentioned target product to print the product title report corresponding to the above-mentioned future product turnover, and dynamically adjust the component position of the product display component corresponding to the above-mentioned target product on the product display page according to the above-mentioned future turnover, and control the product warehousing device to dispatch the replenishment machine device to realize the replenishment processing for the above-mentioned future product turnover.

[0012] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0013] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0014] The above-mentioned embodiments of the present disclosure have the following beneficial effects: Through the product replenishment methods of some embodiments of the present disclosure, the deviation rate of product turnover prediction is reduced, and the inventory backlog rate and out-of-stock rate under the traditional replenishment model are optimized. Specifically, the reasons for the high deviation rate of product turnover prediction, inventory backlog and out-of-stock in related product replenishment methods are: the lack of historical turnover data for new products, the failure of methods that rely on time series patterns or label training, the deviation of replenishment quantity estimation causing inventory imbalance, and the chaotic product management system that cannot obtain product labels, categories, and the current promotion and sales stage of products. This leads to large deviations in the expert system's prediction based on the coarse dimension of the category, causing the replenishment quantity to deviate from the actual demand. Based on this, the product replenishment method of some embodiments of the present disclosure first obtains a product information set, wherein each product information in the product information set includes a product name and product turnover. Then, each product information in the product information set is preprocessed to generate a smoothed turnover quantity and a product cleaning name, thereby obtaining a smoothed turnover quantity set and a product cleaning name set. This allows product turnover data, along with product labels and category information, to be obtained and cleaned, providing a data foundation for subsequent operations on new products. Secondly, the cleaned product name set and the target product name are input into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector. The target product is the product for which future turnover is to be predicted. By analyzing the dimensional information features between the target product and the product, the information features can be tailored to actual conditions, providing a basis for matching similar products. Next, similarity information is determined between each product name vector in the product name vector set and the target product name vector, resulting in a similarity information set. This similarity information set allows the relationship between the target product and similar products to be determined, and the turnover of similar products can be used as a basis for the target product. Subsequently, based on the smoothed turnover set and the similarity information set, the future product turnover corresponding to the target product is generated. The resulting future product turnover can be used as a basis for the target product turnover, serving as a data basis for replenishment. Finally, the printing device corresponding to the target product is controlled to print the product title report corresponding to the future product turnover volume, and the component position of the product display component corresponding to the target product is dynamically adjusted on the product display page according to the future turnover volume, and the product storage device is controlled to schedule the replenishment machine device to realize the replenishment processing for the future product turnover volume. Therefore, according to the turnover volume of the future product, combined with the replenishment decision and the tool device used, the inventory is cleaned up and increased. In summary, by using the data basis of similar products of the new product, by analyzing the similarity between products to predict the turnover volume of the target product, dynamically adjusting the replenishment strategy, and timely replenishment processing, the rational use of storage resources is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0016] Figure 1 is a flow chart of some embodiments of a product replenishment method according to the present disclosure;

[0017] Figure 2 is a schematic structural diagram of some embodiments of the product replenishment device according to the present disclosure;

[0018] Figure 3 It is a structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0019] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0020] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0021] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0022] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0023] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0024] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0025] refer to Figure 1, shows a process 100 of some embodiments of the product replenishment method according to the present disclosure. The product replenishment method may include the following steps:

[0026] Step 101: Obtain product information set.

[0027] In some embodiments, the execution entity (e.g., an electronic device) of the above-mentioned product replenishment method may be a computing terminal storing multiple pieces of product data information. The acquisition method may be wireless or wireless. The product information may be product content describing the target circulating product. In practice, the product information may include: product name and product circulation volume. The target circulating product may be a product for which product circulation volume forecasting is to be performed. In a retail scenario, the target circulating product may be an item for which sales volume forecasting is to be performed. The product information may be the item name. The product circulation volume may be the historical sales data of the item.

[0028] Step 102 : Pre-process each product information in the above product information set to generate a smoothed flow volume and a product cleaning name, thereby obtaining a smoothed flow volume set and a product cleaning name set.

[0029] In some embodiments, the execution entity may pre-process each product information in the product information set to generate a smoothed flow volume and a product cleaning name, thereby obtaining a smoothed flow volume set and a product cleaning name set. The smoothed flow volume is the flow volume data of the product after smoothing and removing outliers. The product cleaning name is the product name after cleaning and removing irrelevant information.

[0030] As an example, the execution entity can smooth the smoothed flow volume, replacing the smoothed flow volume data at the current location with the smoothed flow data corresponding to the previous moment. The execution entity can also cleanse product names by counting information with specific patterns in the product name data. These patterns can include Chinese spaces, Chinese commas, Chinese brackets, and other non-semantic symbols. A classifier trained using a machine learning algorithm removes these patterns from the product names.

[0031] In some optional implementations of some embodiments, the execution entity may pre-process each product information in the product information set to generate a smoothed flow volume and a product cleaning name, which may include the following steps:

[0032] The first step is to determine the mean and standard deviation of the turnover volume based on the product turnover volume included in the above product information. The mean turnover volume can be the mean turnover volume of the product turnover volume, and the standard deviation turnover volume can be the standard deviation of the turnover volume of the product turnover volume. For example, the product turnover volume can be (10, 20, 30), then the mean turnover volume is 20, and the standard deviation turnover volume is 8.16. In a retail scenario, the mean turnover volume can be the mean sales volume of historical sales. The standard deviation turnover volume can be the standard deviation of sales volume of historical sales.

[0033] The second step is to determine the lower limit value and the upper limit value of the inner limit interval based on the above-mentioned turnover volume mean and the above-mentioned turnover volume standard deviation. Among them, the inner limit interval can be a normal turnover volume range interval, which is a boundary for data screening that excludes outliers. In a retail scenario, the inner limit interval can be a normal sales volume range interval, which is a boundary for data screening that excludes sales volume outliers. For example, the lower limit value of the inner limit interval is the first quartile. The first quartile can be the turnover volume mean plus the first multiple value. The lower limit value of the inner limit interval is the third quartile. The third quartile can be the turnover volume mean minus the first multiple value. The first multiple value can be a real number multiple of the turnover volume standard deviation. For example, the real number can be 0.6745.

[0034] In step three, in response to the product flow rate being less than the lower limit, the product flow rate is corrected to the lower limit. For example, the product flow rate may be 20 and the lower limit may be 50. For example, if the product flow rate is 20 and less than the lower limit of 50, the product flow rate is corrected to the lower limit of 20.

[0035] In step 4, in response to the product flow rate being greater than the upper limit, the product information value is corrected to the upper limit. For example, the product flow rate value may be 120, and the upper limit may be 80. For example, if the product flow rate value of 120 is greater than the upper limit of 80, the product flow rate value is corrected to the upper limit of 80.

[0036] The fifth step is to determine the corrected product flow volume as the smoothed flow volume. For example, the product flow volume value may be 120, and after correction it becomes 80, then 80 is determined as the product flow volume.

[0037] The sixth step is to determine the regular text features corresponding to the product names included in the product information. The regular text features can be fixed symbol combinations used for structured segmentation in each product information, such as Chinese spaces, Chinese commas, and Chinese brackets, which are semantically insignificant symbols.

[0038] The seventh step is to perform content cleaning on the regular text features to obtain the cleaned product name. In practice, content cleaning of regular text features can be to remove some fixed symbols in the product name to achieve the generation of pure text. As an example, the execution subject can perform regular expression-based targeted elimination on the regular feature content. For example, "Apple Red Fuji (1 box) gift box" is cleaned into "Apple Red Fuji 1 box gift box".

[0039] Step 103: Input the above-mentioned product cleaning name set and target product name into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector.

[0040] In some embodiments, the execution entity inputs the product cleaning name set and the target product name into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector. The product cleaning name set may be the name of the product after cleaning. The target product name may be the name of the target product for which product turnover is to be predicted. In practice, for retail scenarios, the target product may be a new product. The corresponding target product name may be the name of a newly launched new product. The pre-trained model may be an algorithmic model that numerically represents the natural language form of the product name. The product name vector may be information in vector form that represents the semantics of the name content corresponding to the product name. The target product name vector may be information in vector form that represents the semantics of the name content corresponding to the new product.

[0041] In practice, a pre-trained model can create an index for each unique word or character and use the index as input to construct a binary vector to represent the numerical value. For example, the pre-trained model can be one of the following: N-gram statistical algorithm model, One-Hot encoding model.

[0042] In some optional implementations of some embodiments, the execution entity may input the product cleaning name set and the target product name into a pre-trained model for product name vector conversion to obtain the product name vector set and the target product name vector, which may include the following steps:

[0043] In the first step, for each product cleaning name in the above product cleaning name set, perform the following generation steps:

[0044] The first sub-step is to segment the product cleaning name to obtain product segmentation information, wherein the product segmentation information is the product cleaning name character string segmented into independent words or character strings.

[0045] In practice, you can use a dictionary-based word segmentation tool to segment the product name. For example, the name of a product cleaning product can be "Apple Red Fuji 1 box gift box", and the product segmentation information can be ["apple", "Red Fuji", "1 box", "gift box"].

[0046] The second sub-step is to extract keywords from the above-mentioned product segmentation information. The above-mentioned keywords are words that reflect the main attributes of the product. First, the analysis in the product information is traversed, and the preset standard library is matched to determine the match. The preset standard library can be a standard brand library, a standard category library, or a standard specification library. Then, the matched words are used as keywords. For example, the product segmentation information can be ["apple", "red", "Red Fuji", "Fuji", "1 box", "gift box", "gift box"]. The main attributes of the product can be brand words and category words, and the corresponding keywords can be "apple" and "Red Fuji".

[0047] The third sub-step involves performing entity recognition on the product segmentation information to obtain product name entities. The product name entities can be entity categories identified and labeled with specific meanings from unstructured text. The unstructured text can be product segmentation information. For example, entity categories can be brand, category, or specification.

[0048] In practice, an entity recognition model can generate entity results by converting each text sequence into a sequence of entity labels. For example, the entity recognition model can be one of the following: a large language model, a CRF (Conditional Random Field) or other sequence labeling-specific models.

[0049] In retail scenarios, the entity recognition model can use the BIOES (Begin, Inside, Outside, End, Single) labeling system, covering e-commerce product datasets with brand, category, and specification entities. The Hugging FaceTransformers library can be fine-tuned to obtain entities that can recognize segmented information. For example, the labeling system can be in the form of "B-Brand" or "I-Spec." The segmented information can be ["New*land", "A2β-casein", "whole milk", "6 bottles"], and the entity recognition result can be brand (New*land), category (whole milk), and specification (6 bottles).

[0050] The fourth sub-step is to determine the entity relationship corresponding to the product name entity as the semantic relationship result. These entity relationships can include the relationship between category and brand, and the relationship between specification and category. For example, the relationship between category and brand can be a subordinate relationship. For example, "Brand X" belongs to the "Smartphone" category, and "Fuji" belongs to the "Fruit" category.

[0051] The fifth sub-step is to mine deep semantic content based on the above keywords and the above semantic relationship results to obtain product semantic information. Among them, the above deep semantic content can be high-level semantic features extracted from keywords and entity relationships. The above product semantic information can be a set of structured business attributes to accurately describe business value and application scenarios. As an example, the product name can be "Red Fuji Apple 200g Gift Box". The basic attributes can be brand (Red Fuji), category (apple), specification (200g gift box). The deep semantic content can be market positioning (high-end), consumption scenario (gift), function (organic certification). Product semantic information can be a logical association between entities. For example, market positioning-brand (high-end, Red Fuji).

[0052] In practice, combining the aforementioned keywords and semantic relationship structures, through rule engines and deep semantic reasoning on knowledge graphs, a structured set of business attributes is generated. Knowledge graphs can be industry knowledge bases, such as organic certification lists and brand awareness rankings. Rule engine design can prioritize explicit rules over model reasoning. For example, "gift box" can be combined with "gift."

[0053] For example, a keyword might be {brand: ["Red Fuji"], category: ["apple"], specification: ["200g gift box"]}. The semantic relationship structure might be [brand-category (Red Fuji, apple), category-specification (apple, 200g gift box)]. Through deep semantic reasoning, we can derive the deep semantics corresponding to the basic attributes {brand: ["Red Fuji"], category: ["apple"], specification: ["200g gift box"]}: {market positioning: ["high-end"], consumption scenario: ["gift"], functional characteristics: ["organic certification"], logical association: [brand-category: ("Red Fuji", "apple"), category-specification-scenario: ("apple", "200g gift box", "gift scenario")]}.

[0054] The sixth sub-step involves comparing the aforementioned product semantic information with the target product name to determine the differences in emphasis on brand, category, and specification, thereby obtaining the product's key features. The aforementioned differences in emphasis can be comparative weights for brand, category, specification, and packaging. For example, a product name might be "ZZ Cherries, JJ Grade, 5-Jan Package." The target product name might be "LL Kiwi, 1-Jan Package, Golden Fruit Gift Box." Differences in emphasis can be the weights of "Cherries" and "Kiwis" in terms of category, "5-Jan Package" and "1-Jan Package" in terms of specification, and "No Gift Box" and "Golden Fruit Gift Box" in terms of packaging.

[0055] In practice, multi-dimensional semantic similarity calculations and weighting strategies are used to quantify dimensional differences. Dimensions can be brand, category, and size. Weighting strategies can prioritize matching products with the same brand, followed by products with similar categories.

[0056] For example, the semantic information of a target product might be the basic attributes {Brand: ["Red Fuji"], Category: ["Apple"], Specification: ["200g Gift Box"]}. Deeper semantics might be {Market Positioning: ["High-End"], Consumption Scenario: ["Gift"]}. Product semantic information might be the basic attributes {Brand: ["Telunsu"], Category: ["Milk"], Specification: ["12-Pack"]}. Deeper semantics might be {Market Positioning: ["Mass Market"], Consumption Scenario: ["Family"]}. For example, the degree of difference analysis might be: Brand: 0, Category: 0, Specification: 0.1. The brands are completely different, the categories have no overlap, and the capacity and packaging differ significantly.

[0057] For example, the semantic information of a target product can be represented by the basic attributes {brand: ["Red Fuji"], category: ["Apple"], size: ["200g gift box"]}. The deeper semantics can be represented by {market positioning: ["high-end"], consumption scenario: ["gift"]}. The semantic information of a product can be represented by the basic attributes {brand: ["Aksu"], category: ["Apple"], size: ["500g bulk"]}. The deeper semantics can be represented by {market positioning: ["mass"], consumption scenario: ["family"]}. For example, the degree of difference analysis can be represented by brand: 0, category: 1, size: 0.4. The brands are completely different, the categories are the same, but the capacity and packaging are different.

[0058] In the seventh sub-step, the product cleaning name is simplified based on the key product features to generate a key cleaning name. The key product features may be features that differentiate the product cleaning name from the target product name. The key cleaning name may be the product key name with redundant information removed. For example, the product cleaning name may be "Apple Red Fuji 1 Box Gift Box," while the key cleaning name may be "Red Fuji Gift Box," retaining the category and specification features.

[0059] In practice, a dynamic cleaning rule engine is used to prioritize key product features, retaining key words for core terms like brand, category, and specification to generate standardized, streamlined names. Dynamic cleaning rules can be used to retain brands, simplify categories, and optimize specifications based on specific product characteristics.

[0060] As an example, the original name of the product can be "Xinjiang Aksu Vitamin Nutrition Apple Family Edition 500g Bulk", retaining the brand "Aksu", simplifying the category, deleting "Vitamin Nutrition", and optimizing the specifications from "Family Edition 500g Bulk" to "500g Bulk".

[0061] The second step is to input the resulting set of product cleaning key names into the pre-trained model for product name vector conversion to obtain a set of product name vectors. The pre-trained model for product name vector conversion can be the Transformer encoder. Fine-tuning the Transformer encoder using existing data can yield a model for converting natural language into vectors.

[0062] In a retail scenario, a pre-trained model can be one that uses an e-commerce product name dataset to fine-tune the Transformer encoder model and encode product names into 768-dimensional vectors.

[0063] In the third step, the target product name is input into the pre-trained model for product name vector conversion to obtain the target product name vector. For example, if the product name is "LL Kiwi 1 catties Golden Fruit Gift Box", the target product name can be [0.82, -0.12, 0.43, 1.22, -0.05, ..., 0.91].

[0064] Step 104 : Determine similarity information between each product name vector in the product name vector set and the target product name vector to obtain a similarity information set.

[0065] In some embodiments, the execution entity may determine similarity between each product name vector in the product name vector set and the target product name vector, thereby obtaining a similarity information set. The similarity information may include the degree of semantic match between the two product vectors in terms of brand, category, and specification. In practice, cosine similarity can be used to measure the degree of vector similarity.

[0066] In some optional implementations of some embodiments, the execution entity determining similarity information between each product name vector in the product name vector set and the target product name vector may include the following steps:

[0067] In the first step, the product name vector and the target name vector are dimensionally compressed to obtain a compressed product name vector and a compressed target name vector. The compressed product name vector can be a new vector obtained by reducing the dimensionality of the original product name vector while preserving as much semantic information as possible. The compressed target name vector can be a new vector obtained by reducing the dimensionality of the target name vector.

[0068] In practice, the execution entity performs vector compression on the product name vector and the target name vector, mapping the high-dimensional vectors to a lower-dimensional space. For example, a high-dimensional vector with 768 dimensions is mapped to a lower-dimensional space with 64 dimensions. Using principal component analysis (PCA), the principal component analysis method (PCA) is used to preserve 95% of the original data's variance, reducing computational complexity.

[0069] The second step is to determine the Euclidean distance between the compressed product name vector and the compressed target name vector as similarity information. The Euclidean distance can be used to determine the geometric distance between two vectors in space.

[0070] In practice, the execution entity may calculate the Euclidean distance between the compressed vector of the product name and the compressed vector of the target name, and convert the Euclidean distance into a readable similarity score. The similarity score is determined as similarity information. The similarity score may be a real number minus a first ratio, where the first ratio may be the ratio of the Euclidean distance to the maximum distance. As an example, the real number may be 1, and the maximum distance may be 5. For example, a distance of 3.0 corresponds to a similarity score of 1-3.0 / 5.0=0.4.

[0071] Step 105: Generate the future product turnover corresponding to the target product based on the smoothed turnover set and the similarity information set.

[0072] In some embodiments, the execution entity generates a future product turnover volume corresponding to the target product based on the smoothed turnover volume set and the similarity information set. The future product turnover volume may be the target product's turnover volume in a future time period. In retail scenarios, the future product turnover volume may be the sales volume of a new product in a future time period. The future product turnover volume may be a forecast of the target product's inventory turnover demand based on historical product data, for replenishment decisions.

[0073] In practice, the above-mentioned execution entity can generate future product turnover through a weighted average. First, similar products are weighted, such as by brand, category, or specification. Then, the historical turnover of similar products is normalized and weighted averaged according to the weights. Finally, the final prediction value is calculated by multiplying the weighted average result with the target product benchmark. The target product benchmark can be the maximum historical turnover of the product with the highest similarity information.

[0074] For example, for product A in the similarity information set, the similarity information can be 0.8, and the turnover of product A in a period of time in the smoothed turnover set can be [50, 55, 60]. For product B in the similarity information set, the similarity information can be 0.6, and the turnover of product B in a period of time in the smoothed turnover set can be [40, 45, 50]. After normalization, the turnover of products A and B is [0.83, 0.92, 1.0] and [0.80, 0.90, 1.0], respectively. The normalization can be the ratio of the turnover in a period of time to the maximum turnover. The weighted calculation result can be 0.8 / (0.8+0.6)×[0.83, 0.92, 1.0]+0.6 / (0.8+0.6)×[0.80, 0.90, 1.0]=[0.82, 0.91, 1.0]. The future product turnover can be 60×[0.82, 0.91, 1.0]=[49, 55, 60].

[0075] In some optional implementations of some embodiments, the execution entity may generate the future product turnover corresponding to the target product based on the smoothed turnover set and the similarity information set, which may include the following steps:

[0076] The first step is to normalize each similar information in the above similar information set to generate a corresponding flow weight matrix. Among them, the above flow weight matrix can be a matrix of normalized similar information, which can be used to quantify the contribution of each similar product to the target prediction. As an example, there can be three products in the similar information set, product A, product B, and product C, and the corresponding similarities are 0.3, 0.6, and 0.9 respectively. Then it can be determined that the minimum similarity is 0.3, the maximum similarity is 0.9, and the difference is 0.9-0.3=0.6. Then perform the normalization operation, product A is normalized to (0.3-0.3) / 0.6=0.0, product B is normalized to (0.6-0.3) / 0.6=0.5, and product C is normalized to (0.9-0.3) / 0.6=1.0. Then the normalized flow weight matrix can be [0.0, 0.5, 1.0].

[0077] In practice, the execution entity performs a normalization operation on the similarity information, and uses the ratio of the first difference value to the second difference value to map the similarity score to the interval [0, 1]. The first difference value can be the value of the current similarity score minus the minimum similarity score. The second difference value can be the maximum similarity score minus the minimum similarity score. As an example, the similarity information set can be [0.2, 0.5, 0.8], which will be normalized to [0.0, 0.5, 1.0].

[0078] The second step is to perform matrix multiplication and weighted averaging on the aforementioned flow weight matrix and the aforementioned smoothed flow set to obtain the initial target product flow. The external features mentioned above include: target flow label, market index, and seasonal factor. The seasonal factor is based on the growth rate of historical data over the same period. The market index is a dynamic indicator reflecting overall market demand.

[0079] In a retail scenario, the target flow tag can be a promotional tag. The seasonal factor can be the growth rate during the forecast period. For example, if sales on the first day of the current cycle are 500 and 400 on the first day of the same period last year, the seasonal factor is 1.25. If sales on the second day of the current cycle are 400 and 200 on the second day of the same period last year, the seasonal factor is 2.00. The market index can be constructed based on real-time product flow data, industry reports, or macroeconomic data. For example, the market index can be a market trend index. In a retail scenario, the market index can be a trend value determined by e-commerce platform search volume or social media popularity.

[0080] The third step is to perform a second correction on the initial target product turnover volume based on external characteristics to obtain the final future product turnover volume. In retail scenarios, this initial target turnover volume reflects the weighted average of similar historical products and does not take into account the dynamic impact of external characteristics. Therefore, a second correction is required to incorporate external characteristics.

[0081] The fourth step is to determine the final future product turnover as the future product turnover. The final future product turnover is a predicted value corrected by external characteristics and can be used to guide inventory replenishment scheduling.

[0082] While adopting technical solutions to address the problem of replenishing new product launches, the following issues often arise: Lack of adaptability to the market environment: Relying solely on historical data on similar products without considering current market dynamics, such as seasonal variations and regional consumption differences. Lagging real-time demand response: Historical data is unable to address unexpected situations. For example, social media promotions can lead to competing products being out of stock. The immediate shifts in demand for new products can make replenishment strategies incapable of responding to market changes.

[0083] Conventional solutions to these problems typically rely on manual experience to determine product turnover and schedule replenishment. However, the inventors considered that manual experience-based judgments of product turnover are subject to individual cognitive limitations and lack quantitative data. Manual decision-making processes are lengthy and lack real-time market demand, making it difficult to make timely decisions and impacting the scheduling of new product replenishment. Therefore, we decided to adopt the following solution.

[0084] Optionally, the execution subject performs a second correction on the initial target product turnover volume according to the external characteristics to obtain the final future product turnover volume, which may include the following steps:

[0085] The first step is to convert the target flow label into a binary value to obtain the target flow quantitative value. In a retail scenario, the target flow label can be a promotion status. If the product is on promotion, the binary value is determined to be 1, and if the product is not on promotion, the binary value is 0. For example, the sales of mooncake gift boxes for the Mid-Autumn Festival.

[0086] The second step is to normalize the seasonal factors to obtain quantized seasonal factor values. The seasonal factors are mapped to the interval [0, 1] using the ratio of the first and second differences. The first difference can be the value of the current seasonal factor minus the value of the minimum seasonal factor. The second difference can be the value of the maximum seasonal factor minus the value of the minimum seasonal factor. As an example, the seasonal factors can be [0.5, 1.2, 2.0], which after normalization is [0.0, 0.47, 1.0].

[0087] The third step is to obtain real-time market indices. In practice, real-time market indices can be obtained from external data sources. These external data sources can include databases, web crawlers, or platform APIs (Application Programming Interfaces). For example, the daily search index for a product name can be obtained through an e-commerce platform API. A value of 120 indicates a high-demand product.

[0088] The fourth step is to smooth the real-time market index to obtain smoothed market data. This smoothed market data can be a market index that has been smoothed to reduce noise. In practice, this smoothing process can be performed using a sliding window average method. First, based on the forecast time period, the market index for that period is obtained. Next, the window selection is determined based on the target product cycle, and then the smoothing process is performed.

[0089] For example, the original real-time market index may be [100, 110, 120], and the window may be 3. In the first period,

[100] is smoothed to 100. In the second period, [100, 110] is smoothed to 105. In the third period, [100, 110, 120] is smoothed to 110. The real-time market index after smoothing may be [100, 105, 110].

[0090] Step 5: Concatenate the target turnover value, the market smoothing data, and the seasonal factor value to obtain an external feature vector. In practice, first determine the target turnover value, the market smoothing data, and the seasonal factor value. Then, concatenate the different features in a fixed order to form a co-vector. For example, [target turnover value, market smoothing data, seasonal factor value].

[0091] For example, the target turnover value may be 1, the market smoothing data may be 105, and the seasonal factor value may be 0.47. Then the external eigenvector may be [1, 105, 0.47].

[0092] In the sixth step, a pre-trained feature analysis model is used to assign weights to the dimensions of the external feature vector to obtain an external vector. The pre-trained feature analysis model may be a random forest model. The external vector may be a vector of importance scores corresponding to features in the external features. The importance score may be the product of the feature weight and the feature value.

[0093] In practice, this random forest model can be used as an input for external features to determine the relationship between these features and product turnover. The importance of each feature is assessed using node splitting gains, generating a corresponding importance score for each indicator. This score is then used as the feature weight.

[0094] As an example, the weight of the target turnover quantization value can be 0.5. The weight of the market smoothing data can be 0.3. The weight of the seasonal factor quantization value can be 0.2. Then the external eigenvector can be [1, 110, 0.47], and the external vector can be [0.5, 33, 0.094].

[0095] In the seventh step, the external vector and the initial target product turnover are input into a pre-trained gradient boosting decision tree model to obtain the final future product turnover. In a retail scenario, the gradient boosting decision tree can have 100 trees, a learning rate of 0.1, and a maximum depth of 5. The training data can be a historical e-commerce product turnover dataset, and the evaluation metric can be mean squared error. The historical turnover dataset can be data from the past three years.

[0096] Steps 1 to 7, as an inventive feature of the present disclosure, address the third technical problem mentioned in the background art, namely, "In the context of massive products and complex scenarios, the use of manually labeled data or calculations is complex, computationally intensive, and costly, and cannot meet the real-time scheduling requirements for warehouse replenishment." In practice, manual experience is often used to determine the status of goods turnover and schedule replenishment. However, manual experience is easily limited by individual cognition, lacks quantitative data, and cannot be integrated with market demand in real time to make immediate replenishment decisions, which affects the efficiency of new product replenishment scheduling. Therefore, the present disclosure designs a solution based on dynamic correction of external features. The initial target product turnover is corrected based on multi-dimensional external features to achieve new product replenishment scheduling. First, the target turnover label is converted into a binary value to obtain a target turnover quantization value. The external features are numerically quantized to provide a data foundation for the construction of the external vector. Second, the seasonal factor is normalized to obtain a seasonal factor quantization value. Ensure that the impact of seasonal factors in the overall external vector is comparable to that of other external features, thereby increasing the sensitivity of subsequent models to seasonal changes. Next, obtain a real-time market index. Promptly acquiring real-time market demand information allows for rapid perception of changes in market demand for products. This is followed by smoothing the real-time market index to produce smoothed market data. This ensures that the real-time market index reflects factory-wide market demand trends, avoids misjudgments caused by abnormal data, and improves data authenticity. Next, concatenate the target turnover quantified value, the smoothed market data, and the seasonal factor quantified value to produce an external feature vector. This creates a feature vector that includes multi-dimensional information on product promotion status, market demand trends, and seasonal influences, enabling subsequent models to comprehensively consider multiple factors. Then, using a pre-trained feature analysis model, weights are assigned to each dimension in the external feature vector to produce an external vector. This quantifies the contribution of each external feature to product turnover and accurately reflects its impact on product turnover. Finally, the external vector and the initial target product turnover are input into a pre-trained gradient boosting decision tree model to obtain the final future product turnover. By combining multi-dimensional external features with historical turnover data for similar products and gradient-boosting the nonlinear fitting capabilities of the pre-trained decision tree model, we generate target product turnover forecasts that align with actual market demand. This forecast is more closely aligned with actual market demand, providing accurate data support for warehouse replenishment, enabling immediate replenishment scheduling and reducing the risk of stockouts and inventory backlogs.

[0097] In the process of adopting technical solutions to solve the second technical problem mentioned above, the following problems are often accompanied: multi-feature coupling: the high-dimensional product name vector mixes the product's various dimensional information (brand, category, specifications), and does not effectively explain the importance of each dimension, resulting in feature mixing affecting the similarity between products. Lack of business scenario adaptation: Traditional prediction models use fixed weights to determine the importance of each dimension and cannot respond to changes in market focus in different business scenarios. For example, consumers of new products pay attention to the trustworthiness of the brand, and during promotions pay attention to the price-performance ratio of specifications. It is impossible to distinguish which dimensions affect the replenishment decision, resulting in an imbalance in the inventory structure. For example, high-premium brands are out of stock.

[0098] In response to these problems, the conventional solution is generally to extract keywords from product names and match the names of historical similar products to predict the turnover of the target product. The inability to respond to changes in demand in different replenishment scenarios leads to unreasonable replenishment scheduling, resulting in out-of-stock or product inventory backlogs. For example, during a promotion period, users pay attention to the price-performance ratio of specifications and need to increase the replenishment of high-discount specifications. The inventors took into account that extracting keywords from product names and matching the turnover of historical similar products may lead to unclear product selection and ambiguity. For example, a company's product "Apple" is the same as the fruit type "Apple". Using keyword matching to assign a uniform weight ignores the shortcomings of the differences in the importance of dimensions in different business scenarios, and cannot change the weight according to changes in the actual replenishment scenario needs, resulting in deviations in replenishment scheduling. We decided to adopt the following solution.

[0099] In some optional implementations of some embodiments, the execution entity may generate the future product turnover corresponding to the target product based on the smoothed turnover set and the similarity information set, which may include the following steps:

[0100] The first step is to decouple each product name vector in the product name vector set to generate a category vector for the product name vector. The dimensions of the product name vector include brand, category, specification, and semantics. The category vector can be a low-dimensional subvector decoupled from the product name vector.

[0101] In practice, the above-mentioned execution entity can decouple the product name vectors through a multi-head attention model. In retail scenarios, the multi-head attention model can be trained using an e-commerce product name dataset and use cosine similarity to evaluate the decoupling-related model.

[0102] For example, a product name might be "Black Plum Organic Plums 400g Gift Box," and after vectorization, it might be [0.4, ..., 0.7, 0.5, ..., 0.2, 0.7, ..., 0.4]. For example, after decoupling, the brand vector might be [0.4, ..., 0.7], the category vector might be [0.5, ..., 0.2], and the specification vector might be [0.7, ..., 0.4].

[0103] The second step is to decouple the target product name vector to obtain a dimensional feature vector of the target product name. The dimensional feature vector can be a low-dimensional sub-vector decoupled from the target product name vector.

[0104] For example, the target product name could be "Imported Organic Blueberries 200g Gift Box", and after vectorization, it could be [0.8, ..., 0.6, 0.9, ..., 0.5, 0.7, ..., 0.3]. For example, after decoupling, the brand vector could be [0.8, ..., 0.6], the category vector could be [0.9, ..., 0.5], and the specification vector could be [0.7, ..., 0.3].

[0105] The third step is to dynamically adjust the weight coefficients of each dimension based on the target product's business scenario category. Business scenario categories can be phased scenarios based on the product's circulation status at different stages, used to identify the current market's varying levels of reliance on different product attributes. In practice, the current business scenario can be selected based on the actual business status.

[0106] In retail scenarios, business scenarios can be classified into promotional period, regular period, and new product period, and each scenario has a corresponding weighting table. For example, during promotional period, brand weight is 0.7, and specification weight is 0.2; during regular period, category weight is 0.5, and specification weight is 0.4; and during new product period, category weight is 0.3, and brand weight is 0.6.

[0107] The fourth step is to normalize the above weight coefficients to generate a dynamic weight matrix. The dynamic weight matrix can be based on the weight coefficients corresponding to each dimension of the product in different business scenarios, and constrain the original weights in the current scenario.

[0108] For example, the original brand weight can be 0.7, the category weight can be 0.1, and the specification weight can be 0.2. For example, during a promotional period, the brand weight can be 0.7 and the specification weight can be 0.2. The normalized brand weight can be 0.7 / (0.7+0.2)=0.78, and the normalized specification weight can be 0.2 / (0.7+0.2)=0.22. The dynamic weight matrix can then be [0.78, 0.22].

[0109] In the fifth step, dynamically weight the category vectors and dimension feature vectors according to the dynamic weight matrix to obtain a comprehensive similarity information set. This comprehensive similarity information can quantify the differences in the contributions of various attributes to similarity in different business scenarios. Each attribute can be the product attribute represented by the decoupled category vector and dimension feature vector, respectively.

[0110] In practice, the above execution entity performs a dynamic weighted summation operation, and performs a weighted summation of the category vector and the dimension feature vector by matrix multiplication on the dynamic weight matrix. If the dimensions do not match, the broadcast mechanism is used to expand the dimensions. As an example, the target product category vector can be 0.9, the similar product A category vector can be 0.8, and the dynamic weight matrix can be [0.78 (category), 0.22 (brand)]. For example, the comprehensive similarity information set can be (0.78×0.8)+(0.22×0.6)=0.756.

[0111] Step 6: Normalize the above comprehensive similarity information set to generate a similarity information weight matrix. As an example, the comprehensive similarity information set can be [0.756 (Product A), 0.25 (Product B), 0.15 (Product C)]. For example, the total similarity information is 0.756 + 0.25 + 0.15 = 1.156, which is normalized to [0.756 / 1.156, 0.25 / 1.156, 0.15 / 1.156] = [0.65, 0.22, 0.13], and the similarity information weight matrix is [0.65, 0.22, 0.13].

[0112] Step 7: Matrix multiplication is performed on the similarity information weight matrix and the smoothed flow volume set to obtain the final future product flow volume. As an example, the smoothed flow volume set can be Product A: 800, Product B: 200, Product C: 500, and the similarity information weights are [0.65, 0.22, 0.13]. For example, the final future product flow volume is (800 × 0.65) + (200 × 0.22) + (500 × 0.13) = 520 + 44 + 65 = 629.

[0113] Steps 1 through 7, as a key feature of this disclosure, address the second technical issue mentioned in the background technology: "Data chaos in the product management system causes the expert system to predict based on coarse category dimensions, resulting in replenishment volumes deviating from actual demand and an imbalance in warehouse inventory structure." In practice, the turnover of target products is predicted by extracting keywords from product names and matching them with similar historical products. However, keyword matching can lead to ambiguous product selection, and selecting the wrong product can affect subsequent replenishment. Simply relying on keyword matching to assign uniform weights ignores the varying importance of dimensions in different business scenarios, fails to adjust weights based on actual replenishment scenarios, and results in irrational replenishment scheduling. This disclosure designs a solution for dynamically assigning feature weights based on business scenarios. This solution assigns weights to product features according to different business scenarios, accurately predicting the turnover of target products and the replenishment scheduling plan for each business scenario. First, each product name vector in the aforementioned product name vector set is decoupled to generate a category vector for the product name vector. The dimensions corresponding to the product name vector include: brand, category, specification, and semantics. Separating the information from each dimension mixed within the high-dimensional vector avoids feature confounding and provides clear feature data for subsequent similarity calculations. Secondly, the target product name vector is decoupled to obtain the dimensional feature vector for the target product name. This clarifies the representation of each dimensional feature of the target product, providing a data foundation for subsequent weight calculations and similarity assessments. Then, dynamically adjust the weight coefficients for each dimension based on the target product's business scenario category. Based on different business scenarios, the impact of corresponding dimensions on product turnover is highlighted, improving the scenario-specific adaptability of the prediction. Next, normalize the weight coefficients to generate a dynamic weight matrix. This eliminates variations in the weight coefficients, making the weights of each dimension comparable and providing accurate coefficients for subsequent weighted summation. Based on this dynamic weight matrix, a dynamic weighted summation is performed on the category vector and the dimensional feature vector, generating a comprehensive similarity information set. Each dimensional feature is combined with its corresponding weight to obtain a quantitative result that reflects the overall similarity between products. This comprehensive similarity information set is then normalized to generate a similarity information weight matrix. Ensure that the weights in the similarity information weight matrix are within a reasonable range to facilitate calculations based on turnover data. Finally, perform matrix multiplication on the similarity information weight matrix and the smoothed turnover set to obtain the final future product turnover. Combining product similarities with historical turnover data, we generate turnover forecasts for target products in real-world business scenarios and enable timely adjustments to replenishment schedules, mitigating stockouts and inventory backlogs caused by inappropriate replenishment scheduling.

[0114] Step 106: Control the printing device corresponding to the target product to print a product title report corresponding to the future product turnover volume, dynamically adjust the component position of the product display component corresponding to the target product on the product display page according to the future turnover volume, and control the product storage device to dispatch the replenishment machine device to realize the replenishment processing for the future product turnover volume.

[0115] In some embodiments, the execution entity controls the printing device corresponding to the target product to print the product title report corresponding to the future product turnover volume, and dynamically adjusts the component position of the product display component corresponding to the target product on the product display page according to the future turnover volume, and controls the product storage device to dispatch the replenishment machine device to realize the replenishment processing for the future product turnover volume. The product title report can be a summary of the key information of the product presented in the form of text or labels, which is used to intuitively display the replenishment content related to the future product turnover volume. The key information of the product can be basic product information, turnover data, business scenario information, and inventory and replenishment suggestions. The printing device can be a hardware device for outputting the product title report.

[0116] As an example, the printing device may be a printer that receives a product title report and converts it into a paper or label format. For example, the printing device may be a thermal printer.

[0117] In practice, the above-mentioned execution entity dynamically adjusts the product display component, which may be a dynamic adjustment of future circulation volume based on real-time click-through rate data, wherein the dynamic adjustment may be that when the click-through rate is higher than a preset value, the price tag is moved to the corresponding position, and the position calculation may be the current pixel position plus a first difference of 0.5 times, and the first difference may be the click-through rate minus the preset value. For example, the preset value may be 0.1. As an example, the current position may be 100px, the click-through rate may be 0.15, and the moving position may be 100+0.5×(0.15-0.1)=100.25px. In a retail scenario, the above-mentioned click-through rate may be the ratio of the number of clicks to the number of exposures.

[0118] In practice, the aforementioned execution entities can perform replenishment operations by monitoring inventory levels in real time. When inventory levels drop to a preset value, replenishment tasks are generated, and replenishment is determined based on the stock-out risk level. Replenishment tasks can include current inventory levels or safety stock levels. For example, if the stock-out risk is greater than 80%, the aforementioned replenishment machine is immediately assigned to perform replenishment. If the stock-out risk is less than 30%, the task is deferred. In retail scenarios, the stock-out risk can be the percentage of the expected stock-out quantity divided by the average daily sales volume. The expected stock-out quantity can be the target replenishment quantity minus the current inventory level, and the average daily sales volume can be the average daily sales volume over the past 30 days.

[0119] The above-mentioned embodiments of the present disclosure have the following beneficial effects: Through the product replenishment methods of some embodiments of the present disclosure, the deviation rate of product turnover prediction is reduced, and the inventory backlog rate and out-of-stock rate under the traditional replenishment model are optimized. Specifically, the reasons for the high deviation rate of product turnover prediction, inventory backlog and out-of-stock in related product replenishment methods are: the lack of historical turnover data for new products, the failure of methods that rely on time series patterns or label training, the deviation of replenishment quantity estimation causing inventory imbalance, and the chaotic product management system that cannot obtain product labels, categories, and the current promotion and sales stage of products. This leads to large deviations in the expert system's prediction based on the coarse dimension of the category, causing the replenishment quantity to deviate from the actual demand. Based on this, the product replenishment method of some embodiments of the present disclosure first obtains a product information set, wherein each product information in the product information set includes a product name and product turnover. Then, each product information in the product information set is preprocessed to generate a smoothed turnover quantity and a product cleaning name, thereby obtaining a smoothed turnover quantity set and a product cleaning name set. This allows product turnover data, along with product labels and category information, to be obtained and cleaned, providing a data foundation for subsequent operations on new products. Secondly, the cleaned product name set and the target product name are input into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector. The target product is the product for which future turnover is to be predicted. By analyzing the dimensional information features between the target product and the product, the information features can be tailored to actual conditions, providing a basis for matching similar products. Next, similarity information is determined between each product name vector in the product name vector set and the target product name vector, resulting in a similarity information set. This similarity information set allows the relationship between the target product and similar products to be determined, and the turnover of similar products can be used as a basis for the target product. Subsequently, based on the smoothed turnover set and the similarity information set, the future product turnover corresponding to the target product is generated. The resulting future product turnover can be used as a basis for the target product turnover, serving as a data basis for replenishment. Finally, the printing device corresponding to the target product is controlled to print the product title report corresponding to the future product turnover volume, and the component position of the product display component corresponding to the target product is dynamically adjusted on the product display page according to the future turnover volume, and the product storage device is controlled to schedule the replenishment machine device to realize the replenishment processing for the future product turnover volume. Therefore, according to the turnover volume of the future product, combined with the replenishment decision and the tool device used, the inventory is cleaned up and increased. In summary, by using the data basis of similar products of the new product, by analyzing the similarity between products to predict the turnover volume of the target product, dynamically adjusting the replenishment strategy, and timely replenishment processing, the rational use of storage resources is realized.

[0120] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a product replenishment device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the product replenishment device can be specifically applied to various electronic devices.

[0121] like Figure 2 As shown, a product replenishment device 200 includes an acquisition unit 201, a preprocessing unit 202, a vectorization unit 203, a similarity information determination unit 204, a future turnover volume prediction unit 205, and a printing unit 206. The acquisition unit 201 is configured to acquire a product information set, wherein each product information in the product information set includes a product name and a product turnover volume. The preprocessing unit 202 is configured to preprocess each product information in the product information set to generate a smoothed turnover volume and a product cleaning name, thereby obtaining a smoothed turnover volume set and a product cleaning name set. The vectorization unit 203 is configured to input the product cleaning name set and the target product name into a pretrained model for product name vector conversion, thereby obtaining a product name vector set and a target product name vector, wherein the target product is the product for which future turnover volume is to be predicted. The similarity information determination unit 204 is configured to determine similarity between each product name vector in the product name vector set and the target product name vector, thereby obtaining a similarity information set. The future turnover prediction unit 205 is configured to generate the future product turnover corresponding to the target product based on the smoothed turnover set and the similarity information set. The printing device unit 206 is configured to control the printing device corresponding to the target product to print a product title report corresponding to the future product turnover, dynamically adjust the position of the product display component corresponding to the target product on the product display page based on the future turnover, and control the product storage device to schedule the replenishment machine to implement replenishment processing based on the future product turnover.

[0122] It is understood that the units described in the product replenishment device 200 are similar to those in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the product replenishment device 200 and the units included therein, and will not be repeated here.

[0123] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0124] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. Various programs and data required for the operation of the electronic device 300 are also stored in the RAM 303. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0125] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0126] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are performed.

[0127] It should be noted that in some embodiments of the present disclosure, the computer-readable medium mentioned above may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0128] In some embodiments, the client and server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0129] The computer-readable medium may be included in the electronic device; or it may exist independently without being assembled into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a product information set, wherein each product information in the product information set includes a product name and a product turnover; pre-processes each product information in the product information set to generate a smooth turnover and a product cleaning name, and obtains a smooth turnover set and a product cleaning name set; inputs the product cleaning name set and the target product name into a pre-trained model for product name vector conversion, and obtains a product name vector set and a target product name vector, wherein the target product is to be predicted. products with future circulation volume; determining similarity information between each product name vector in the above-mentioned product name vector set and the above-mentioned target product name vector to obtain a similarity information set; generating the future product circulation volume corresponding to the above-mentioned target product based on the above-mentioned smooth circulation volume set and the above-mentioned similarity information set; controlling the printing device corresponding to the above-mentioned target product to print a product title report corresponding to the above-mentioned future product circulation volume, and dynamically adjusting the component position of the product display component corresponding to the above-mentioned target product on the product display page according to the above-mentioned future circulation volume, and controlling the product warehousing device to dispatch the replenishment machine device to realize the replenishment processing for the above-mentioned future product circulation volume.

[0130] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0132] The units described in some embodiments of the present disclosure may be implemented in software or hardware. The units described may also be provided in a processor. For example, they may be described as follows: a processor including an acquisition unit, a preprocessing unit, a vectorization unit, a similarity information determination unit, a future flow volume prediction unit, and a printing device unit. The names of these units do not, in some cases, constitute limitations on the units themselves. For example, the acquisition unit may also be described as a "unit for acquiring a product information set."

[0133] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0134] The above description is only an illustration of some preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A product replenishment method, comprising: Acquire a product information set, wherein each product information in the product information set includes a product name and a product turnover amount; Preprocessing each product information in the product information set to generate a smoothed flow volume and a product cleaning name, thereby obtaining a smoothed flow volume set and a product cleaning name set; Inputting the product cleaning name set and the target product name into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector, wherein the target product is a product for which future turnover is to be predicted; Determine similarity information between each product name vector in the product name vector set and the target product name vector to obtain a similarity information set; generating a future product turnover corresponding to the target product according to the smoothed turnover set and the similarity information set; Control the printing device corresponding to the target product to print a product title report corresponding to the future product turnover volume, and dynamically adjust the component position of the product display component corresponding to the target product on the product display page according to the future turnover volume, and control the product warehousing device to dispatch the replenishment machine device to realize the replenishment processing for the future product turnover volume.

2. The method according to claim 1, wherein The pre-processing of each product information in the product information set to generate a smooth flow volume and a product cleaning name includes: Determine the mean value and standard deviation of the product turnover amount according to the product turnover amount included in the product information; Determining a lower limit value of an inner limit interval and an upper limit value of an inner limit interval according to the turnover amount mean and the turnover amount standard deviation; In response to the value of the product flow rate being less than the lower limit value, correcting the value of the product flow rate to the lower limit value; In response to the value of the product flow rate being greater than the upper limit, correcting the value of the product information to the upper limit; Determine the corrected product flow volume as the smoothed flow volume; Determining regular text features corresponding to the product name included in the product information; Content cleaning is performed on the regular text features to obtain the product cleaning name.

3. The method according to claim 1, wherein: Determining similarity information between each product name vector in the product name vector set and the target product name vector includes: Performing vector dimension compression on the product name vector and the target name vector to obtain a product name compressed vector and a target name compressed vector; The Euclidean distance between the compressed product name vector and the compressed target name vector is determined as similarity information.

4. The method according to claim 1, wherein: Generating the future product turnover corresponding to the target product according to the smoothed turnover set and the similarity information set includes: Normalizing each piece of similar information in the similar information set to generate a corresponding flow weight matrix; Performing matrix multiplication and weighted averaging on the flow volume weight matrix and the smoothed flow volume set to obtain an initial target product flow volume; Performing a second correction on the initial target product turnover volume according to the external characteristics to obtain a final future product turnover volume; The final future product turnover volume is determined as the future product turnover volume.

5. The method according to claim 1, wherein The product cleaning name set and the target product name are input into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector, including: For each product cleaning name in the product cleaning name set, perform the following generation steps: Segmenting the product cleaning name to obtain product segmentation information; Extracting key words from the product segmentation information; Perform entity recognition on the product segmentation information to obtain a product name entity; Determine the entity relationship corresponding to the product name entity as the semantic relationship result; Mining deep semantic content based on the keywords and the semantic relationship results to obtain product semantic information; Compare the product semantic information with the target product name to determine the differences in emphasis on brand, category, and specifications, and obtain the key features of the product; Simplify the product cleaning name according to the key features of the product to generate the product cleaning key name; Inputting the obtained product cleaning key name set into the pre-trained model for product name vector conversion to obtain a product name vector set; The target product name is input into the pre-trained model for product name vector conversion to obtain a target product name vector.

6. A product replenishment device comprising: an acquiring unit configured to acquire a product information set, wherein each product information in the product information set includes a product name and a product turnover amount; a preprocessing unit configured to preprocess each product information in the product information set to generate a smoothed flow volume and a product cleaning name, thereby obtaining a smoothed flow volume set and a product cleaning name set; a vectorization unit configured to input the product cleaning name set and the target product name into a pre-trained model for product name vector conversion to obtain a product name vector set and a target product name vector, wherein the target product is a product for which future circulation volume is to be predicted; A similarity information determining unit is configured to determine similarity information between each product name vector in the product name vector set and the target product name vector to obtain a similarity information set; a future circulation volume prediction unit configured to generate a future product circulation volume corresponding to the target product based on the smoothed circulation volume set and the similarity information set; The printing device unit is configured to control the printing device corresponding to the target product to print a product title report corresponding to the future product turnover volume, and dynamically adjust the component position of the product display component corresponding to the target product on the product display page according to the future turnover volume, and control the product warehousing device to dispatch the replenishment machine device to realize the replenishment processing for the future product turnover volume.

7. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images