本文關(guān)鍵詞： 深度學(xué)習(xí) 農(nóng)產(chǎn)品銷量預(yù)測(cè) 農(nóng)產(chǎn)品銷量評(píng)價(jià)指標(biāo) ICM　出處：《計(jì)算機(jī)應(yīng)用研究》2017年08期 　論文類型：期刊論文

【摘要】：針對(duì)線上農(nóng)產(chǎn)品銷售存在的信息不對(duì)稱問題,提出一種結(jié)合深度學(xué)習(xí)算法優(yōu)勢(shì)和涉農(nóng)電商銷售數(shù)據(jù)特點(diǎn)的皇冠模型(ICM)。首先建立因素評(píng)價(jià)指標(biāo),將銷量分為四個(gè)類別;其次采用兩層自編碼網(wǎng)絡(luò)提取樣本特征,并生成新的特征向量;然后利用帶標(biāo)簽樣本集訓(xùn)練分類器并對(duì)無標(biāo)簽訓(xùn)練樣本分類;最后利用BP微調(diào)整個(gè)網(wǎng)絡(luò)參數(shù)得到使損失函數(shù)值達(dá)到最小的最優(yōu)參數(shù),實(shí)現(xiàn)線上農(nóng)產(chǎn)品的銷量分類預(yù)測(cè)。經(jīng)仿真分析,驗(yàn)證了ICM的分類準(zhǔn)確率高達(dá)88%,明顯高于其他未將數(shù)據(jù)進(jìn)行特征學(xué)習(xí)的淺層分類器,證明了ICM具有較好的增量自學(xué)習(xí)能力和層次認(rèn)知能力。
[Abstract]:Aiming at the problem of information asymmetry existing in online agricultural product sales, a crown model combining the advantages of depth learning algorithm and the characteristics of sales data of agribusiness is proposed. Firstly, the evaluation index of factors is established, and the sales volume is divided into four categories. Secondly, a two-layer self-coding network is used to extract the sample features and generate a new feature vector, and then the untagged training samples are classified by using the labeled sample set to train the classifier and to classify the untagged training samples. Finally, the optimal parameters of the loss function are obtained by fine-tuning the whole network parameters, and the classification and prediction of the sales volume of agricultural products on line are realized. It is verified that the classification accuracy of ICM is as high as 88%, which is obviously higher than that of other shallow classifiers without feature learning. It is proved that ICM has better incremental self-learning ability and hierarchical cognitive ability.
【作者單位】： 河北工業(yè)大學(xué)計(jì)算機(jī)科學(xué)與軟件學(xué)院;河北省大數(shù)據(jù)計(jì)算重點(diǎn)實(shí)驗(yàn)室;河北工業(yè)大學(xué)經(jīng)濟(jì)管理學(xué)院;
【基金】：天津市軟科學(xué)基金項(xiàng)目(16450303D) 河北省社會(huì)科學(xué)基金資助項(xiàng)目(HB15GL112) 河北省科技計(jì)劃資助項(xiàng)目(16450303D)
【分類號(hào)】：F323.7;F724.6;TP181
，

本文編號(hào)：1502187