TWI828426B - An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts - Google Patents
An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts Download PDFInfo
- Publication number
- TWI828426B TWI828426B TW111143808A TW111143808A TWI828426B TW I828426 B TWI828426 B TW I828426B TW 111143808 A TW111143808 A TW 111143808A TW 111143808 A TW111143808 A TW 111143808A TW I828426 B TWI828426 B TW I828426B
- Authority
- TW
- Taiwan
- Prior art keywords
- product
- sales
- keywords
- module
- keyword
- Prior art date
Links
- 238000000605 extraction Methods 0.000 claims abstract description 17
- 238000010801 machine learning Methods 0.000 claims abstract description 16
- 238000001514 detection method Methods 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000009193 crawling Effects 0.000 abstract 1
- 238000007726 management method Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 7
- 238000013277 forecasting method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000013016 learning Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本發明係為一種利用多面向商品情蒐資訊進行銷售預測之電商輔助系統，主要利用商品情蒐模組所收集彙整資訊供銷售預測模組機器學習後進行銷售預測。其中該商品情蒐模組包含：商品狀態模組，供呈現商品上架與庫存資訊；銷售動能偵測模組，供呈現每一商品於一區間內多個不同之銷售動能表現；關鍵字擷取模組，透過使用者訂單取得複數個普遍性關鍵字，並於該等普遍性關鍵字選取對應各商品之代表性關鍵字並予以標記；關鍵字通路排名模組，爬取各通路平台之熱門關鍵字；以及關鍵字社群排名模組，根據代表性關鍵字爬取網路論壇內容，據此依爬取結果更新各商品標記之代表性關鍵字。The invention is an e-commerce auxiliary system that uses multi-oriented product information search information to perform sales forecasting. It mainly uses the information collected and compiled by the product information search module for machine learning by the sales prediction module to make sales forecasts. The product information search module includes: a product status module for displaying product shelf and inventory information; a sales momentum detection module for displaying multiple different sales momentum performances of each product within a range; keyword extraction The module obtains a plurality of universal keywords through user orders, and selects representative keywords corresponding to each product from these universal keywords and tags them; the keyword channel ranking module crawls the popular keywords of each channel platform. Keywords; and the keyword community ranking module, which crawls online forum content based on representative keywords, and updates the representative keywords of each product tag based on the crawling results.
Description
本發明係與網路購物之電商領域有關，特別是一種利用多面向商品情蒐資訊進行銷售預測之電商輔助系統。 The present invention relates to the e-commerce field of online shopping, and is particularly an e-commerce auxiliary system that uses multi-oriented product search information to conduct sales predictions.
商品銷售策略一直以來無論線下或線上，眾所周知地其大原則不外乎係販售銷量較佳之熱門商品，以及增加熱門商品庫存量等基本市場供需概念。然，縱使知其道理，業者們該如何正確判斷各通路之銷售情形，以及各種商品之銷售趨勢等，目前多數仍採用個別資訊評估而由決策單位依過去經驗累積進行判斷。然而現今電商活動交易頻繁且變化度極高，在此環境基礎下本即難以快速交叉分析各種銷售變因及影響比重，且由於數據資訊本身即屬抽象，是以所相應之隱含銷售關係更非可輕易讀取或認知，因此實務上最後常流於主觀判斷或依直覺進行決策。另方面，以落後資訊作為決策依據，因其本質即帶有已反應之指標特性，故對於未來之預測趨勢更常有失精準，業者們亦難依此作為中短程商品之上架與庫存有效配置。 Product sales strategies, whether offline or online, have always been based on basic market supply and demand concepts such as selling popular products with higher sales volume and increasing the inventory of popular products. However, even if they know the truth, how do businesses correctly judge the sales situation of each channel and the sales trends of various commodities? Currently, most of them still use individual information evaluation and make judgments based on the accumulation of past experience by decision-making units. However, today's e-commerce activities have frequent transactions and extremely high variability. Under this environment, it is difficult to quickly cross-analyze various sales variables and impact proportions. Moreover, because the data information itself is abstract, the corresponding implicit sales relationship It cannot be easily read or recognized, so in practice, decision-making often ends up relying on subjective judgment or intuition. On the other hand, using backward information as the basis for decision-making is often inaccurate in predicting future trends because of its nature of reflected indicators. It is also difficult for operators to rely on this to effectively allocate short- and medium-range products on the shelves and inventory. .
鑒於前述缺失，先前技術已不乏有提出各種面向之銷售預測的相關專利文獻，例如公開號CN15555025A之銷售預測管理系統、方法及記錄介質，係利用歷史銷售紀錄與用戶輸入數據進行未來銷售預測。惟該文獻充其量僅不過把既有人為分析方法予以系統化，對於所謂之分析模型如何建立或運作並無多所著墨。另公開號CN102214338A之銷售預測系統及方法，亦是維持利用歷史資料為預測基底之基礎，並設計數據、統計、分類、分解及期間滾動技術，以 期建立科學化之預測模型。惟，具體上該些供以建模之數據資訊意義與內涵為何並無進一步具體揭示。更進一步地，後續亦不乏有如公開號CN108256924A之一種產品銷售預測裝置；公開號CN110599234A之一種產品銷售預測方法；公開號CN112381591A之基於LSTM深度學習模型的銷售預測優化方法；公開號CN11419412A之基於智慧商品系統銷售預測的自動補貨方法或補貨系統；公開號114240483A之一種基於機器學習及提前分類的新零售終端銷售預測方法與系統；公開號CN114387037A之基於深度學習的零售商品銷售預測方法；公開號CN114971687之一種基於Attention機制的LSTM-BP組合模型的銷售預測方法；公開號CN115147153A之一種基於分層建模的商品銷售預測系統和方法；公開號TW201837814A之一種無模型推測基礎的產品銷售預測方法及系統；公開號TW202113698A之銷售預測系統與方法等，係為各種利用既有之基本機器學習技術，而將各式不同取向下所定義之資料內容供訓練建模再為後續預測之技術手段，故如何摘取並處理各式學習訓練資料始為前述機器學習重點。換言之，縱使對於未來之事本即存在不可預測性，但對於前述該些預測方法或系統，其終端目的仍是欲取得相對較為精準之預測結果，而該等技術之實質差異，不外乎係利用提供訓練之資料處理、屬性內容、清洗程度、資訊定義，分類方式等面向不同，強調據此供機器學習後生成不同層次與處理對象或相異領域之各式預測模型具有較佳預測功效。例如，公開號CN107292672A之一種餐飲行業銷售預測的實現系統與方法；或公開號TW202109383A之圖書銷售預測設備及其預測方法，係進一步輔以爬蟲與關鍵字等技術供以建立預測系統。 In view of the above deficiencies, there is no shortage of related patent documents proposing various aspects of sales forecasting in the prior art. For example, the sales forecast management system, method and recording medium of Publication No. CN15555025A uses historical sales records and user input data to conduct future sales forecasts. However, this literature only systematizes existing artificial analysis methods at best, and does not say much about how the so-called analysis model is established or operated. In addition, the sales forecasting system and method of Publication No. CN102214338A also maintains the use of historical data as the basis for forecasting, and designs data, statistics, classification, decomposition and period rolling technologies to A scientific prediction model will be established in the future. However, the specific meaning and connotation of the data information used for modeling has not been further revealed in detail. Furthermore, there are many follow-up publications such as a product sales forecasting device with Publication No. CN108256924A; a product sales forecasting method with Publication No. CN110599234A; a sales forecast optimization method based on LSTM deep learning model with Publication No. CN112381591A; and a smart product-based publicity No. CN11419412A. Automatic replenishment method or replenishment system for system sales forecast; Publication number 114240483A, a new retail terminal sales forecasting method and system based on machine learning and early classification; Publication number CN114387037A, a retail commodity sales forecasting method based on deep learning; Publication number CN114387037A CN114971687, a sales forecasting method based on the LSTM-BP combination model based on the Attention mechanism; Publication No. CN115147153A, a commodity sales forecasting system and method based on hierarchical modeling; Publication No. TW201837814A, a product sales forecasting method without model speculation basis and System; the sales forecasting system and method of Publication No. TW202113698A is a technical means that utilizes existing basic machine learning technologies to use data content defined in various orientations for training modeling and subsequent prediction. Therefore, How to extract and process various learning and training data is the focus of the aforementioned machine learning. In other words, even though there is inherent unpredictability in the future, the ultimate goal of the aforementioned prediction methods or systems is still to obtain relatively accurate prediction results, and the essential difference between these technologies is nothing more than the Utilizing different aspects of data processing, attribute content, cleaning degree, information definition, and classification methods to provide training, it is emphasized that various prediction models based on this for machine learning to generate different levels and processing objects or different fields have better prediction effects. For example, the publication number CN107292672A is a system and method for implementing sales forecasting in the catering industry; or the publication number TW202109383A is a book sales forecasting device and its prediction method, which are further supplemented by technologies such as crawlers and keywords to establish a forecasting system.
承上所述，眾所周知地在大數據環境時代下，機器學習、爬蟲技術、關鍵字比對等皆屬數據分析、統計、建模等基礎技術，惟如何利用前述技 術針對各個領域提出並建立較佳學習模型始為本案欲以改善之重點。有感於此，考量現今網路電子購物之電商領域係為大眾消費主流，而在充斥多種電商平台以及賣家與產品多元性考量下，如何讓業者得以有效管控各式商品之進銷存資訊，以及讓各式商品與數量如何適得其所地分別上架於各個通路，而讓總體營銷創造最大值並讓庫存風險拿捏亦得有所依存，故本團隊提出一種利用多面向商品情蒐資訊進行銷售預測之電商輔助系統，藉此供各業者得以在為前述各方面執行時得有所適從與依歸，讓電商決策不再盲目行之。 As mentioned above, it is well known that in the era of big data environment, machine learning, crawler technology, keyword comparison, etc. are all basic technologies such as data analysis, statistics, and modeling. However, how to use the aforementioned technologies? The focus of this project is to propose and establish better learning models for various fields. In view of this, considering that the e-commerce field of online e-shopping is now the mainstream of mass consumption, and considering the variety of e-commerce platforms and the diversity of sellers and products, how can the industry effectively control the purchase, sale and inventory of various goods? Information, as well as how to put various products and quantities on the shelves in each channel in the right place, which also depends on creating maximum value for overall marketing and controlling inventory risks. Therefore, our team proposes a method to use multi-oriented product information search The e-commerce auxiliary system for sales forecasting based on information provides various operators with appropriate guidance and guidance when executing the above-mentioned aspects, so that e-commerce decisions are no longer made blindly.
本發明之主要目的在於提供一種可透過電腦系統運算以獲悉在現時時點及近程未來市場中各商品之預期銷售情況，進而藉此有效配置於各電商平台賣場之各式商品種類、數量以及倉儲管控之銷售預測系統。 The main purpose of the present invention is to provide a method that can be used to calculate the expected sales situation of each commodity in the current time point and the near-future market through computer system calculations, and thereby effectively allocate the types, quantities, and types of various commodities in each e-commerce platform store. Sales forecasting system for warehouse management and control.
為實現上述目的，本發明係揭露一種利用多面向商品情蒐資訊進行銷售預測之電商輔助系統，包含：一商品情蒐模組，電訊連接至少一電商伺服器，包含：一通路銷售模組，包含：一商品狀態模組，供以調取及管理該電商伺服器中各式商品於至少一通路平台之上架與庫存資訊；及一銷售動能偵測模組，對應該電商伺服器中各商品而依日期區間與過去銷量經計算而於同一商品在同一日期區間內分別生成複數個銷售動能；及一商品代表字匹配模組，電訊連接該電商伺服器，包含：一關鍵字擷取模組，透過該電商伺服器之複數個使用者訂單計算取得每一商品之複數個普遍性關鍵字及其權重，並於該等普遍性關鍵字中選取權重較高者定義為至少一代表性關鍵字，再將該代表性關鍵字標記於對應之商品；一關鍵字通路排名模組，爬取並彙整各該通路平台之至少一熱門關鍵字，對比該等普遍性關鍵字與該熱門關鍵字以調整該等普遍性關鍵字權重；及一關鍵字社群排名模組，根據該等普遍性關鍵字爬取至少一網路論 壇之相關內容以調整該等普遍性關鍵字權重；其中，該商品代表字匹配模組依據該關鍵字通路排名模組與該關鍵字社群排名模組之權重調整結果更新各商品之該代表性關鍵字並標記；及一銷售預測模組，電訊連接該商品情蒐模組，並根據一訓練資料透過機器學習以對銷售量進行預測；其中該訓練資料為前述各式商品於該通路平台之上架與庫存資訊、該等銷售動能及該代表性關鍵字。 In order to achieve the above object, the present invention discloses an e-commerce auxiliary system for sales prediction using multi-oriented product information search information, including: a product information search module, telecommunications connected to at least one e-commerce server, including: a channel sales module. The group includes: a product status module for retrieving and managing the shelf and inventory information of various products in at least one channel platform in the e-commerce server; and a sales momentum detection module corresponding to the e-commerce server Each product in the device is calculated based on the date range and past sales volume to generate multiple sales momentum for the same product within the same date range; and a product representative word matching module is connected to the e-commerce server via telecommunications, including: a key The word extraction module calculates and obtains multiple universal keywords and their weights for each product through multiple user orders of the e-commerce server, and selects the higher weighted keyword among these universal keywords to define as At least one representative keyword, and then mark the representative keyword on the corresponding product; a keyword channel ranking module, crawl and aggregate at least one popular keyword of each channel platform, and compare the common keywords and the popular keywords to adjust the weight of the universal keywords; and a keyword community ranking module to crawl at least one online discussion based on the universal keywords The relevant content of the forum is used to adjust the weight of these universal keywords; wherein, the product representative word matching module updates the representative of each product based on the weight adjustment results of the keyword channel ranking module and the keyword community ranking module. and tag the specific keywords; and a sales forecast module, which is connected to the product information search module via telecommunications, and predicts sales volume through machine learning based on a training data; wherein the training data is the aforementioned various products on the channel platform Shelf and inventory information, such sales momentum, and such representative keywords.
較佳地，其中該銷售動能係利用最小二乘法針對日期區間與過去銷量計算所獲得，且該等銷售動能係以複數個N日趨勢線呈現。進一步地，為使每一商品各自地分別呈現短、中、長區間意義之該等銷售動能，其中該等N日趨勢線可為3日趨勢線、7日趨勢線、及30日趨勢線之該銷售動能。 Preferably, the sales momentum is obtained by calculating the date interval and past sales volume using the least squares method, and the sales momentum is presented as a plurality of N-day trend lines. Further, in order for each commodity to respectively present the sales momentum in the short, medium and long range, the N-day trend lines may be the 3-day trend line, the 7-day trend line, and the 30-day trend line. That sales momentum.
較佳地，其中該關鍵字擷取模組更包含：一白名單資訊與一黑名單資訊，供以該關鍵字擷取模組透過該電商伺服器之該等使用者訂單計算取得該代表性關鍵字過程，過濾剔除屬於該白名單資訊與該黑名單資訊之字詞。藉此過濾不必要之贅詞以提高該代表性關鍵字之擷取精準度。 Preferably, the keyword extraction module further includes: a whitelist information and a blacklist information, for the keyword extraction module to calculate and obtain the representative through the user orders of the e-commerce server The specific keyword process filters out words belonging to the whitelist information and the blacklist information. This will filter out unnecessary redundant words to improve the accuracy of retrieving the representative keywords.
較佳地，其中，該關鍵字擷取模組係以TF-IDF計算取得該等普遍性關鍵字及對應權重，且該等普遍性關鍵字對應之IDF，係由總該等使用者訂單資訊文件數目除以包含該等普遍性關鍵字之文件數目，再將得到的商取以3為底的對數計算取得。 Preferably, the keyword retrieval module calculates the universal keywords and corresponding weights using TF-IDF, and the IDF corresponding to the universal keywords is calculated from the total user order information. The number of documents is divided by the number of documents containing the universal keywords, and the resulting quotient is calculated by taking the logarithm of base 3.
較佳地，其中該關鍵字社群排名模組係依據該網路論壇中關於該等普遍性關鍵字討論之一熱度頻率調整權重。進一步地，該關鍵字社群排名模組更包含：一正向語意資訊庫與一負向語意資訊庫，當該關鍵字社群排名模組針對該網路論壇中關於該等普遍性關鍵字討論之相關語意進行分析後，根據渉及之正向語意及負向語意數量調整該等普遍性關鍵字權重。 Preferably, the keyword community ranking module adjusts the weight according to the popularity frequency of discussions about the common keywords in the online forum. Further, the keyword community ranking module further includes: a positive semantic information database and a negative semantic information database. When the keyword community ranking module targets the universal keywords in the online forum, After analyzing the relevant semantics of the discussion, the weight of these universal keywords is adjusted according to the number of positive and negative semantics involved.
較佳地，其中該銷售預測模組，係以前述任一實施例所據此獲取之多面向商品情蒐資訊做為該訓練資料，再輔以類神經網路LSTM為機器學習而對銷售量進行預測。 Preferably, the sales prediction module uses multi-oriented product information search information obtained in any of the above embodiments as the training data, and is supplemented by a neural network LSTM for machine learning to predict the sales volume. Make predictions.
較佳地，其中，該商品狀態模組更具有一SKU管控單元，供以執行組合貨號組合綁定與出貨贈品設定。進一步地，該商品狀態模組更生成一商品管理介面，供以快速調整並呈現各商品之安全庫量與可接單量，以及匯入SKU、批次更新及新增SKU作動。 Preferably, the product status module further has a SKU control unit for performing combination binding of product numbers and setting of shipping gifts. Furthermore, the product status module generates a product management interface for quickly adjusting and displaying the safe inventory and available order quantities of each product, as well as importing SKUs, batch updates, and adding SKU actions.
綜上所述，本發明係透過多面向之商品情蒐資訊，再透過機器學習而對銷售量及相應之庫存量進行預測，由於該商品情蒐資訊包含了計算所得各商品之數個銷售動能資訊，因此透過銷售動能可即時反應商品庫存是否應予調整，例如以不同時間區段為基礎之不同N日趨勢線交叉時點作為商品庫存量調整判斷依據。再者，利用商品訂單之描述性資訊，擷取對應該商品之較佳代表性關鍵字，進一步與各通路排名以及網路社群討論熱度或正負面評價進行對比後調整各商品之代表性關鍵字，據此加深並即時反應於現時狀態下商品之銷售情形，而由於該些商品情蒐資訊間之互涉關聯性極高，因此本發明之預測模型無需大量資料而可僅利用該些學習資料即得針對大趨勢進行預測，後續供業者依此預測結果有效管控各種商品上架與否與合理庫存之較佳配置。 To sum up, the present invention predicts sales volume and corresponding inventory through multi-faceted product search information and machine learning, because the product search information contains several calculated sales momentum of each product. Information, therefore, sales momentum can be used to immediately reflect whether product inventory should be adjusted. For example, different N-day trend line crossing points based on different time periods can be used as the basis for judgment of product inventory adjustment. Furthermore, the descriptive information of the product order is used to extract the better representative keywords corresponding to the product, and the representative keywords of each product are adjusted after further comparing with each channel ranking and the popularity of online community discussions or positive and negative reviews. Based on this, we can deepen and immediately reflect the sales situation of the product in the current state. Since the correlation between the product information is extremely high, the prediction model of the present invention does not require a large amount of data and can only use these learnings. The data can then be used to predict major trends, and subsequent suppliers can effectively control the availability of various products on the shelves and the optimal allocation of reasonable inventory based on the prediction results.
1:商品情蒐模組 1: Product information search module
10:通路銷售模組 10: Channel sales module
101:商品狀態模組 101: Product status module
1011:SKU管控單元 1011:SKU control unit
1012:商品管理介面 1012: Product management interface
102:銷售動能偵測模組 102:Selling kinetic energy detection module
12:商品代表字匹配模組 12: Product representative word matching module
121:關鍵字擷取模組 121:Keyword extraction module
122:關鍵字通路排名模組 122:Keyword channel ranking module
123:關鍵字社群排名模組 123:Keyword community ranking module
161:正向語意資訊庫 161: Forward semantic information database
162:負向語意資訊庫 162: Negative semantic information database
2:銷售預測模組 2: Sales forecast module
3:電商伺服器 3: E-commerce server
30:通路平台 30: Access platform
第1圖，為本發明較佳實施例之系統功能方塊圖。 Figure 1 is a system functional block diagram of a preferred embodiment of the present invention.
第2A圖，為本發明較佳實施例之銷售動能以3日趨勢線與7日趨勢線對比圖。 Figure 2A is a comparison chart of sales momentum based on the 3-day trend line and the 7-day trend line according to the preferred embodiment of the present invention.
第2B圖，為本發明較佳實施例之銷售動能以7日趨勢線與30日趨勢線對比圖。 Figure 2B is a comparison chart of sales momentum based on the 7-day trend line and the 30-day trend line according to the preferred embodiment of the present invention.
第3圖，為本發明較佳實施例之代表性關鍵字擷取流程圖。 Figure 3 is a flow chart of representative keyword extraction according to the preferred embodiment of the present invention.
第4圖，為本發明較佳實施例之普遍性關鍵字與各通路排名資訊比對以調整權重流程圖。 Figure 4 is a flow chart of comparing universal keywords with each channel ranking information to adjust weights according to the preferred embodiment of the present invention.
第5圖，為本發明較佳實施例之普遍性關鍵字與各社群網路討論程度比對以調整權重流程圖。 Figure 5 is a flow chart for adjusting weights by comparing universal keywords with the discussion levels of various social networks according to a preferred embodiment of the present invention.
第6圖，為本發明較佳實施例之銷售暨庫存預測趨勢圖。 Figure 6 is a sales and inventory forecast trend chart according to the preferred embodiment of the present invention.
第7圖，為本發明較佳實施例之商品管理介面示意圖。 Figure 7 is a schematic diagram of the product management interface of the preferred embodiment of the present invention.
為使本領域具有通常知識者能清楚了解本發明之內容，謹以下列說明搭配圖式，敬請參閱。 In order to enable those with ordinary knowledge in the art to clearly understand the contents of the present invention, the following description is accompanied by the drawings, for which please refer.
請參閱第1及7圖，其係為本發明較佳實施例之系統功能方塊圖及商品管理介面示意圖。如圖所示，本發明係揭露一種利用多面向商品情蒐資訊進行銷售預測之電商輔助系統，包含：一商品情蒐模組1及一銷售預測模組2，且銷售預測模組2電訊連接該商品情蒐模組1，並根據一訓練資料透過機器學習以對銷售量進行預測。其中該商品情蒐模組1電訊連接至少一電商伺服器3。該電商伺服器3可為一個或多個電腦所組成之伺服器叢集以做為後台運作設備，並據此架構及生成對應之前台通路供業者或消費者進行電商消費行為。進一步地，該商品情蒐模組1包含：一通路銷售模組10及一商品代表字匹配模組12。其中該通路銷售模組10係供業者紀錄、查詢並調取針對目前業者之所有訂單及商品相關資訊進行資訊收集及回饋。例如各式賣場之SKU單位管理或分類，商品可接單量、已接單量、價格、庫存量、出貨管理等資訊。其中該通路銷售模組10包含：一商品狀態模組101及一銷售動能偵測模組102。且該商品狀態模組101，供以調取及管理該電商伺服器3中各式商品於至少一通路平台30之上架與庫存 資訊。一般來說，縱使同一廠商業者亦可能同時將商品上架於不同之該通路平台30，例如MOMO、PCHOME或蝦皮同時上架該業者所提供之商品。因此透過該商品狀態模組101可總覽性的將該些賣場之商品資訊予以紀錄及查詢回饋。進一步地，該商品狀態模組101更具有一SKU管控單元1011，供以執行組合貨號組合綁定與出貨贈品設定；且該商品狀態模組101更生成一商品管理介面1012，供以快速調整並呈現各商品之安全庫量與可接單量，以及匯入SKU、批次更新及新增SKU作動。 Please refer to Figures 1 and 7, which are system functional block diagrams and product management interface schematic diagrams of preferred embodiments of the present invention. As shown in the figure, the present invention discloses an e-commerce auxiliary system for sales forecasting using multi-oriented product information search information, including: a product information search module 1 and a sales forecast module 2, and the sales forecast module 2 is telecommunications Connect the product information search module 1, and use machine learning to predict sales volume based on a training data. The product information search module 1 is connected to at least one e-commerce server 3 via telecommunications. The e-commerce server 3 can be a server cluster composed of one or more computers as a back-end operation device, and based on this structure and generation of corresponding front-end channel providers or consumers for e-commerce consumption behavior. Further, the product information search module 1 includes: a channel sales module 10 and a product representative word matching module 12. Among them, the channel sales module 10 is for the supplier to record, query and retrieve all orders and product-related information for the current supplier for information collection and feedback. For example, SKU unit management or classification of various stores, information such as the number of orders that can be received, the number of orders received, price, inventory, and shipping management. The channel sales module 10 includes: a product status module 101 and a sales momentum detection module 102. And the product status module 101 is used to retrieve and manage the shelving and inventory of various products in the e-commerce server 3 on at least one channel platform 30 information. Generally speaking, even the same manufacturer may put products on different channel platforms 30 at the same time. For example, MOMO, PCHOME or Shopee may put the products provided by the same manufacturer at the same time. Therefore, through the product status module 101, the product information of these stores can be recorded and inquired back in an overview. Furthermore, the product status module 101 further has a SKU control unit 1011 for performing combination binding of product numbers and setting of shipping gifts; and the product status module 101 further generates a product management interface 1012 for quick adjustment. It also displays the safety inventory and available order quantity of each product, as well as the actions of importing SKU, batch update and adding SKU.
接續地，該銷售動能偵測模組102，係對應該電商伺服器3中各商品而依日期區間與過去銷量經計算而於同一商品在同一日期區間內分別生成複數個銷售動能。由於每一商品在一定日期區間內本有其各自存在之銷售情況，而該銷售情況在過去而言，係難以透過具銷售意義之量化數據予以對應說明。例如過去僅知商品之各日銷量，但總體而言並無實際量化可參酌之趨勢數值。因此本發明係提出並定義一種全新概念之該等銷售動能，藉此供以表現銷售量與之前相比是否有顯著的降低或是成長。例如某一該通路平台之衛生紙前日銷量為300包今日為310包；而雨衣前日銷量為5件，但由於昨日天氣預報豪雨將至，故今日銷量提升至50件，雖雨衣總銷售量沒有衛生紙多，但成長確有顯著提升，故該日期區間內雨衣對應之該等銷售動能，理論上將比衛生紙對應之該等銷售動能有明顯提升，而該等銷售動能即本發明創設之數值處理結果並供作機器學習之該訓練材料的其中資訊之一。附帶一提地，透過該等銷售動能表現亦可即時供業者知悉何種商品的庫存應予以增加或調整。進一步地，該銷售動能係可利用最小二乘法針對日期區間與過去銷量計算通過最小化誤差的平方和尋找數據的最佳函數匹配所獲得，且該等銷售動能最終係以複數個N日趨勢線呈現。而為使每一商品可各自地分別呈現短、中、長區間意義之該等銷售動能，其中該等N日趨勢線可為3日趨勢線、7日趨勢線、及30日趨勢線之該銷售動能表 現。又為了因應各種產品之本質差異性所導致之該銷售動能應以不同區間觀察始能更進一步貼近實際銷售狀況，因此該N日趨勢線亦可經數值化處理後呈現為60日、90日、120日或更長區間之該銷售動能，例如高單價與低單價商品、或長固商品與消耗品等所對應之該銷售動能反應波動程度其伴隨之區間表現自然不同。是以，針對不同商品仍有選擇較佳且適應於該商品個性之多個該銷售動能之必要，但基本上透過3、7、30日仍可大致呈現多數商品之短中長之該銷售動能。請一併參閱第2A圖及第2B圖所示，係分別為本發明較佳實施例之銷售動能以3日趨勢線與7日趨勢線對比圖，及銷售動能以7日趨勢線與30日趨勢線對比圖。解讀方式通常可以股票之不同日K線的黃金交叉或死亡交差概念為相同判斷邏輯。如以第2A圖為例，若三天的動能值即將超過七天動能值，則代表該商品持續被關注，越陡峭則關注者越多；另觀第2B圖，其中若七天的動能值開始低於三十天動能值，則該商品熱度即開始消失。因此透過該等銷售動能，係可依此初判某一商品銷售之趨勢方向。再次說明地，每一商品在某一區間本即存在中性之銷售量資訊，而本發明係針對同一商品在相同日期區間下，以既有之銷售數據相比不同時間區段為基底而計算獲得之多個N日線趨勢線的銷售動能。此與過去一般銷售預測系統僅是利用某一時間區間的對應銷量作為歷史比對，作為該訓練資料之數值對應意義本案仍與先前技術作法明顯不同。 Subsequently, the sales momentum detection module 102 is calculated according to the date interval and past sales volume corresponding to each product in the e-commerce server 3 to generate a plurality of sales momentum for the same product within the same date interval. Since each product has its own sales situation within a certain date range, this sales situation has been difficult to explain correspondingly through quantitative data with sales significance in the past. For example, in the past, we only knew the daily sales volume of products, but overall there was no actual quantitative trend value for reference. Therefore, the present invention proposes and defines a brand new concept of sales momentum, which is used to express whether sales volume has significantly decreased or increased compared with before. For example, the sales volume of toilet paper on a certain channel platform was 300 packs the day before and 310 packs today; and the sales volume of raincoats the day before was 5 pieces. However, due to the heavy rain forecast yesterday, the sales volume today increased to 50 pieces. Although the total sales volume of raincoats is not that of toilet paper However, the growth has indeed increased significantly. Therefore, the sales momentum corresponding to the raincoat within the date range will theoretically be significantly higher than the sales momentum corresponding to the toilet paper, and the sales momentum is the numerical processing result created by the present invention. And provide one of the information as training materials for machine learning. Incidentally, these sales momentum performances can also provide businesses with real-time information on which products should be increased or adjusted in inventory. Furthermore, the sales momentum can be obtained by using the least squares method to calculate the best function match of the data by minimizing the sum of squares of errors for the date range and past sales, and the sales momentum is finally based on a plurality of N-day trend lines. Present. In order to allow each commodity to separately present the sales momentum in the short, medium and long range, the N-day trend lines can be the 3-day trend line, the 7-day trend line, and the 30-day trend line. sales momentum meter now. In order to cope with the essential differences of various products, the sales momentum should be observed in different intervals to get closer to the actual sales situation. Therefore, the N-day trend line can also be numerically processed and presented as 60-day, 90-day, The sales momentum in the 120-day or longer range, such as high unit price and low unit price products, or long-term goods and consumables, etc., corresponds to the fluctuation degree of the sales momentum response and the accompanying range performance is naturally different. Therefore, it is still necessary to select multiple sales momentum for different products that are better and suitable for the characteristics of the product. However, basically, the short, medium and long sales momentum of most products can still be roughly displayed through the 3rd, 7th and 30th. . Please refer to Figure 2A and Figure 2B together, which are comparison charts of the sales momentum with the 3-day trend line and the 7-day trend line, and the sales momentum with the 7-day trend line and the 30-day trend line respectively according to the preferred embodiment of the present invention. Trend line comparison chart. The interpretation method can usually be based on the concept of golden cross or death cross of different daily K-lines of stocks as the same judgment logic. Take Figure 2A as an example. If the three-day kinetic energy value is about to exceed the seven-day kinetic energy value, it means that the commodity continues to be paid attention to. The steeper it is, the more followers it has. Looking at Figure 2B, if the seven-day kinetic energy value starts to decrease. At the 30-day kinetic energy value, the popularity of the commodity begins to disappear. Therefore, through these sales momentum, we can initially judge the trend direction of a certain product's sales. To illustrate again, each product already has neutral sales volume information in a certain interval, and this invention is based on the calculation of the same product in the same date interval, based on the existing sales data in different time intervals. Gain sales momentum from multiple N-day trend lines. This is different from the past general sales forecasting system that only uses the corresponding sales volume in a certain time period as a historical comparison. The numerical corresponding meaning of the training data is still significantly different from the previous technical approach.
接續地，該訓練資料中之其他資訊內容係透過該商品代表字匹配模組12予以蒐集提供，該商品代表字匹配模組12電訊連接該電商伺服器3，包含：一關鍵字擷取模組121、一關鍵字通路排名模組122及一關鍵字社群排名模組123。其中該關鍵字擷取模組121，透過該電商伺服器3之複數個使用者訂單計算取得每一商品之複數個普遍性關鍵字及其權重，並於該等普遍性關鍵字中選取權重較高者定義為至少一代表性關鍵字，再將該代表性關鍵字標記於對應之商品。一般而言，針對電子購物平台上之商品本有關於該商品之各式對應特徵 描述，例如商品名稱、型號、特殊規格、附加功能等資訊內容。而為了使後續之該訓練資料內容得以讓機器學習準確地掌握作為商品分類之代表特徵，本發明進一步透過業者目前既有系統中之該等使用者訂單計算取得每一商品之描述資訊，再經運算獲取關於該商品之該等普遍性關鍵字與對應權重。請再一併參閱第3圖，係為本發明較佳實施例之代表性關鍵字擷取流程圖。其中該關鍵字擷取模組121更包含：一白名單資訊與一黑名單資訊，供以該關鍵字擷取模組121透過該電商伺服器3之該等使用者訂單計算取得該代表性關鍵字過程，過濾剔除屬於該白名單資訊與該黑名單資訊之字詞。藉此可過濾不必要之贅詞以提高最終取得該代表性關鍵字之擷取精準度。運算上，該關鍵字擷取模組121係以TF-IDF(Term Frequency-Inverse Document Frequency)計算取得該等普遍性關鍵字及對應權重，同時考量電商商品之描述性文字相對較短，故該等普遍性關鍵字對應之IDF，係由總該等使用者訂單資訊文件數目除以包含該等普遍性關鍵字之文件數目，再將得到的商取以3為底的對數計算取得。因此，透過該關鍵字擷取模組121分析取得使用者訂單之商品資訊後，利用斷詞及刪除預設之常見白名單資訊與黑名單資訊剔除贅字，透過TF-IDF取得該等普遍性關鍵字及對應權重，最後再依較高權重之普遍性關鍵字選擇並定義作為符合描述該商品之一個或數個之該代表性關鍵字而予以標記在該商品。 Subsequently, other information content in the training data is collected and provided through the product representative word matching module 12. The product representative word matching module 12 is connected to the e-commerce server 3 via telecommunications and includes: a keyword extraction module. Group 121, a keyword channel ranking module 122 and a keyword community ranking module 123. The keyword retrieval module 121 calculates and obtains a plurality of universal keywords and their weights for each product through a plurality of user orders of the e-commerce server 3, and selects the weights among these universal keywords. The higher one is defined as at least one representative keyword, and then the representative keyword is marked on the corresponding product. Generally speaking, for products on electronic shopping platforms, there are various corresponding characteristics of the products. Description, such as product name, model, special specifications, additional functions and other information. In order to enable the subsequent training data content to allow machine learning to accurately grasp the representative characteristics of product classification, the present invention further obtains the description information of each product through calculation of user orders in the industry's current existing system, and then The calculation is performed to obtain the general keywords and corresponding weights for the product. Please refer to Figure 3 again, which is a representative keyword extraction flow chart of a preferred embodiment of the present invention. The keyword extraction module 121 further includes: a white list information and a black list information, for the keyword extraction module 121 to calculate and obtain the representativeness through the user orders of the e-commerce server 3 The keyword process filters out words belonging to the whitelist information and the blacklist information. This can filter out unnecessary redundant words to improve the accuracy of extracting the representative keywords. In terms of operation, the keyword extraction module 121 calculates the universal keywords and corresponding weights using TF-IDF (Term Frequency-Inverse Document Frequency), and also considers that the descriptive text of e-commerce products is relatively short, so The IDF corresponding to the universal keywords is calculated by dividing the total number of user order information documents by the number of documents containing the universal keywords, and then taking the base 3 logarithm of the resulting quotient. Therefore, after analyzing and obtaining the product information of the user's order through the keyword extraction module 121, we use word segmentation and delete the default common whitelist information and blacklist information to remove redundant words, and obtain the universality through TF-IDF Keywords and corresponding weights are finally selected based on the universal keywords with higher weights and defined as one or several representative keywords that match the description of the product and marked on the product.
請再一併參閱第4及5圖，係為本發明較佳實施例之普遍性關鍵字與各通路排名資訊比對以調整權重流程圖，及普遍性關鍵字與各社群網路討論程度比對以調整權重流程圖。為了使該代表性關鍵字得以真實反應於現實各該通路平台30之排名表現，以及現實該商品於各大論壇之討論熱度，本發明更利用該關鍵字通路排名模組122，爬取各該通路平台30之通路資訊而彙整至少一熱門關鍵字，並對比該等普遍性關鍵字與該熱門關鍵字以調整原先取得之該等普 遍性關鍵字權重，藉此將經調整後權重相對較低而屬退流行之該代表性關鍵字，或經調整後權重相對較高而屬遺漏之該代表性關鍵字予以更新；以及利用該關鍵字社群排名模組123，根據該等普遍性關鍵字爬取至少一網路論壇之相關內容以調整該等普遍性關鍵字權重；其中該關鍵字社群排名模組123係依據該網路論壇中關於該等普遍性關鍵字討論之一熱度頻率調整權重。進一步地，該關鍵字社群排名模組123更包含：一正向語意資訊庫161與一負向語意資訊庫162，當該關鍵字社群排名模組123針對該網路論壇中關於該等普遍性關鍵字討論之相關語意進行分析後，根據涉及之正向語意及負向語意數量調整該等普遍性關鍵字權重。 Please refer to Figures 4 and 5 again, which is a flow chart of comparing universal keywords with each channel ranking information to adjust the weight, and comparing universal keywords with the discussion level of each social network according to the preferred embodiment of the present invention. To adjust the weight flow chart. In order to enable the representative keywords to truly reflect the ranking performance of each channel platform 30 and the discussion popularity of the product in major forums, the present invention further utilizes the keyword channel ranking module 122 to crawl each channel Compile at least one popular keyword based on the channel information of the channel platform 30, and compare the general keywords with the popular keywords to adjust the originally obtained general keywords. Universal keyword weights, whereby the representative keywords whose adjusted weights are relatively low and are out of popularity, or the representative keywords whose adjusted weights are relatively high and which are missing are updated; and use the The keyword community ranking module 123 crawls the relevant content of at least one online forum based on the universal keywords to adjust the weight of the universal keywords; the keyword community ranking module 123 is based on the website Adjust the weight of the popularity frequency of one of the discussions on these common keywords in the road forum. Further, the keyword community ranking module 123 further includes: a positive semantic information database 161 and a negative semantic information database 162. When the keyword community ranking module 123 targets the relevant information in the online forum, After analyzing the relevant semantics of universal keyword discussions, the weights of these universal keywords are adjusted based on the number of positive and negative semantics involved.
承此，該商品代表字匹配模組12則再依據該關鍵字通路排名模組122與該關鍵字社群排名模組123之權重調整結果更新各商品之該代表性關鍵字並予以標記。接續地，該銷售預測模組2即可利用前述各式商品於該通路平台之上架與庫存資訊、該等銷售動能及該代表性關鍵字資訊內容作為該訓練資料，進一步輔以例如類神經網路LSTM等各式演算法為機器學習而對銷售量進行預測。請再一併參閱第6圖，係為本發明較佳實施例之銷售暨庫存預測趨勢圖。由圖所示，該預測特性係無須大量資料即可呈現大趨勢之預測表現，同時對於週期性之線圖表現亦同作為預測考量。 Accordingly, the product representative word matching module 12 then updates the representative keywords of each product and tags them based on the weight adjustment results of the keyword channel ranking module 122 and the keyword community ranking module 123 . Subsequently, the sales forecast module 2 can use the above-mentioned various products on the channel platform and inventory information, the sales momentum and the representative keyword information as the training data, further supplemented by, for example, a neural network. Various algorithms such as LSTM are used to predict sales volume for machine learning. Please refer to Figure 6 again, which is a sales and inventory forecast trend chart of the preferred embodiment of the present invention. As shown in the figure, this prediction feature can predict the performance of the general trend without requiring a large amount of data. At the same time, the performance of cyclical line charts can also be considered for prediction.
綜上所述，本發明係透過多面向之商品情蒐資訊，再透過機器學習而對銷售量及相應之庫存量進行預測，由於該商品情蒐資訊包含了計算所得各商品之數個銷售動能資訊，因此透過銷售動能可即時反應商品庫存是否應予調整，例如以不同時間區段為基礎之不同N日趨勢線交叉時點作為商品庫存量調整判斷依據。再者，利用商品訂單之描述性資訊，擷取對應該商品之較佳代表 性關鍵字，進一步與各通路排名以及網路社群討論熱度或正負面評價進行對比後調整各商品之代表性關鍵字，據此加深並即時反應於現時狀態下商品之銷售情形，而由於該些商品情蒐資訊間之互涉關聯性極高，因此本發明之預測模型無需大量資料而可僅利用該些學習資料即得針對大趨勢進行預測，後續供業者依此預測結果有效管控各種商品上架與否與合理庫存之較佳配置。 To sum up, the present invention predicts sales volume and corresponding inventory through multi-faceted product search information and machine learning, because the product search information contains several calculated sales momentum of each product. Information, therefore, sales momentum can be used to immediately reflect whether product inventory should be adjusted. For example, different N-day trend line crossing points based on different time periods can be used as the basis for judgment of product inventory adjustment. Furthermore, use the descriptive information of the product order to extract a better representative of the product. The representative keywords of each product are further compared with the rankings of each channel and the discussion popularity or positive and negative reviews of the online community, and then the representative keywords of each product are adjusted. Based on this, the sales situation of the product in the current state is deepened and immediately reflected, and due to the The correlation between the search information of these products is extremely high. Therefore, the prediction model of the present invention does not require a large amount of data and can only use the learning data to predict the general trend. Subsequent suppliers can effectively control various products based on the prediction results. The best configuration of whether it is on the shelves or not and reasonable inventory.
以上所述者，僅為本發明申請專利範圍中之較佳實施例說明，而非得依此實施例內容據以限定本發明之權利範圍；故在不脫離本發明之均等範圍下所作之文義變化或修飾，仍皆應涵蓋於本發明之申請專利範圍內。 The above is only an illustration of the preferred embodiments within the patentable scope of the present invention, and is not intended to limit the scope of rights of the present invention based on the content of these embodiments; therefore, any changes in the meaning of the present invention may be made without departing from the equal scope of the present invention. or modifications should still be covered by the patentable scope of the present invention.
1:商品情蒐模組 1: Product information search module
10:通路銷售模組 10: Channel sales module
101:商品狀態模組 101: Product status module
1011:SKU管控單元 1011:SKU control unit
102:銷售動能偵測模組 102:Selling kinetic energy detection module
12:商品代表字匹配模組 12: Product representative word matching module
121:關鍵字擷取模組 121:Keyword extraction module
122:關鍵字通路排名模組 122:Keyword channel ranking module
123:關鍵字社群排名模組 123:Keyword community ranking module
161:正向語意資訊庫 161: Forward semantic information database
162:負向語意資訊庫 162: Negative semantic information database
2:銷售預測模組 2: Sales forecast module
3:電商伺服器 3: E-commerce server
30:通路平台 30: Access platform
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW111143808A TWI828426B (en) | 2022-11-16 | 2022-11-16 | An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW111143808A TWI828426B (en) | 2022-11-16 | 2022-11-16 | An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TWI828426B true TWI828426B (en) | 2024-01-01 |
| TW202422439A TW202422439A (en) | 2024-06-01 |
Family
ID=90459043
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW111143808A TWI828426B (en) | 2022-11-16 | 2022-11-16 | An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI828426B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8392228B2 (en) * | 2010-03-24 | 2013-03-05 | One Network Enterprises, Inc. | Computer program product and method for sales forecasting and adjusting a sales forecast |
| CN111401409A (en) * | 2020-02-28 | 2020-07-10 | 创新奇智(青岛)科技有限公司 | Commodity brand feature acquisition method, sales volume prediction method, device and electronic equipment |
| TW202139098A (en) * | 2020-03-31 | 2021-10-16 | 廣達電腦股份有限公司 | Consumption prediction system and consumption prediction method |
| CN113570417A (en) * | 2021-08-09 | 2021-10-29 | 上海明略人工智能(集团)有限公司 | Social digital marketing method and system, storage medium and electronic equipment |
| CN114022204A (en) * | 2021-11-05 | 2022-02-08 | 上海适享文化传播有限公司 | Private retail search hot word prediction and verification method based on social information insight |
| CN115115417A (en) * | 2022-07-25 | 2022-09-27 | 浪潮卓数大数据产业发展有限公司 | Public opinion-based commodity sales data prediction method, equipment and medium |
-
2022
- 2022-11-16 TW TW111143808A patent/TWI828426B/en active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8392228B2 (en) * | 2010-03-24 | 2013-03-05 | One Network Enterprises, Inc. | Computer program product and method for sales forecasting and adjusting a sales forecast |
| CN111401409A (en) * | 2020-02-28 | 2020-07-10 | 创新奇智(青岛)科技有限公司 | Commodity brand feature acquisition method, sales volume prediction method, device and electronic equipment |
| TW202139098A (en) * | 2020-03-31 | 2021-10-16 | 廣達電腦股份有限公司 | Consumption prediction system and consumption prediction method |
| CN113570417A (en) * | 2021-08-09 | 2021-10-29 | 上海明略人工智能(集团)有限公司 | Social digital marketing method and system, storage medium and electronic equipment |
| CN114022204A (en) * | 2021-11-05 | 2022-02-08 | 上海适享文化传播有限公司 | Private retail search hot word prediction and verification method based on social information insight |
| CN115115417A (en) * | 2022-07-25 | 2022-09-27 | 浪潮卓数大数据产业发展有限公司 | Public opinion-based commodity sales data prediction method, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yadav et al. | Retailer's optimal policy under inflation in fuzzy environment with trade credit | |
| CN103218724B (en) | A kind of system and method that information updating is provided in a communication network | |
| CN109840730B (en) | Method and device for data prediction | |
| Tareq et al. | Framework of dynamic recommendation system for e-shopping | |
| Lingam | The role of Artificial Intelligence (AI) in making accurate stock decisions in E-commerce industry | |
| Yan et al. | Implementation of a product-recommender system in an IoT-based smart shopping using fuzzy logic and apriori algorithm | |
| Ilieva et al. | Big data based system model of electronic commerce | |
| CN112150227A (en) | Commodity recommendation method, system, device and medium | |
| CN113077317A (en) | Item recommendation method, device and equipment based on user data and storage medium | |
| Ahlemeyer-Stubbe et al. | Monetizing Data: How to Uplift Your Business | |
| Satish et al. | Study and Evaluation of user’s behavior in e-commerce Using Data Mining | |
| Ivanov | The digital marketing with the application of cloud technologies | |
| Ivanov | Cloud-based Digital Marketing. | |
| Apichottanakul et al. | Customer behaviour analysis based on buying-data sparsity for multi-category products in pork industry: A hybrid approach | |
| CN116976948A (en) | Method and system for generating dynamic feedback flow diagram of full value chain of manufacturing enterprise | |
| TWI828426B (en) | An e-commerce auxiliary system that uses multi-oriented product search information to conduct sales forecasts | |
| Martasari | Impact of Industrial Technology 4.0 In Improving Service Quality and Customer Experience on E-Commerce Platforms: Literature Review | |
| Chu et al. | Personalized e-commerce website construction based on data mining | |
| Yu et al. | Application of data mining technology in e-commerce | |
| Sheth et al. | A proficient process for Systematic Inventory Management | |
| Fuchs et al. | A stitch in time saves nine-A meta-model for real-time prediction of product returns in ERP systems. | |
| TW202422439A (en) | An e-commerce support system that uses multi-faceted product information collection to predict sales | |
| Shah et al. | Periodic review inventory model with service level constraint in fuzzy environment | |
| JP7458879B2 (en) | Information processing device, information processing method, and information processing program | |
| Valliathal et al. | Designing retailer’s replenishment policies for non-instantaneous deteriorating items under fuzzy costs |