数字信贷产品大数据

数据在金融信贷、产业分析、供应链管理等方面均可适用。使用群体包括但不限于银行、商业保理等领域。解决金融机构获客、企业增信、产业分析，商业保理融资，供应链融资等问题。1.数据预处理、分词:对行政许可数据进行清洗和预处理。包含去除重复数据、填补缺失值、去除停用词和标点符号等，用分词工具对文本进行分词2.使特征提取、构建词典、向量化:使用word2vec算法计算每个词在文本中出现的频率，然后根据频率对词汇进行排序，选取前若干个词汇作为特征，根据特征构建词典,将每个特征映射到一个唯一的整数编号，使用词典将每个行政许可文件转化为向量3.模型训练:使用CRF来训练信息抽取模型;将每个行政许可文件根据其所属行业标记（自定义分类标准）为对应的类别标签，使用神经网络训练行政许可分类器4.信息抽取、许可分类、计算匹配度：使用信息抽取模型，从文本中识别有效期、发证日期、地址、许可证号.使用训练好的行政许可分类器对文件进行分类，并使用余弦相似度（cosθ=A*B/(|A||B|)，A、B是文本词频向量，|A|和|B|分别表示它们的模长，其中A·B表示向量的点积）计算匹配度，最终选取相似度最高的分类作为行政许可类型5.数据标准化：对有效期、发证日期转换成统一的日期格式；对企业名称结合工商数据进行校验，并补充统一社会信用代码等内容对于主体名称疑似有误的情况经过人工介入来校验

This dataset is applicable to financial credit, industrial analysis, supply chain management and other related fields. Its target users include but are not limited to banks, commercial factoring and other relevant sectors. It addresses core issues including customer acquisition for financial institutions, corporate credit enhancement, industrial analysis, commercial factoring financing and supply chain financing.
1. Data Preprocessing and Word Segmentation: Clean and preprocess administrative license data, which includes removing duplicate records, filling missing values, eliminating stop words and punctuation, and conducting word segmentation on the text via a dedicated word segmentation tool.
2. Feature Extraction, Dictionary Construction and Vectorization: Use the Word2Vec algorithm to calculate the frequency of each word in the text, sort the vocabulary based on the calculated frequencies, select the top several words as features, build a dictionary using these features, map each feature to a unique integer ID, and convert each administrative license document into a vector using the constructed dictionary.
3. Model Training: Train the information extraction model using Conditional Random Fields (CRF); tag each administrative license document with corresponding category labels according to its industry affiliation (based on a custom classification standard), and train an administrative license classifier with a neural network.
4. Information Extraction, License Classification and Similarity Calculation: Utilize the trained information extraction model to identify the validity period, issuance date, address and license number from the text. Classify the documents using the pre-trained administrative license classifier, and calculate the similarity with cosine similarity (cosθ = A·B/(|A|×|B|), where A and B are the word frequency vectors of the texts, |A| and |B| represent their respective magnitudes, and A·B denotes the dot product of the two vectors). Finally, select the category with the highest similarity as the administrative license type.
5. Data Standardization: Convert the validity period and issuance date into a unified date format; verify enterprise names against industrial and commercial data, and supplement information such as the Unified Social Credit Identifier (USCI). For cases where the subject name is suspected to be incorrect, perform verification via manual intervention.

搜集汇总

数据集介绍

特点

该数据集为数字信贷产品大数据，包含大量企业行政许可信息，适用于金融信贷和供应链管理等多个领域，每日更新，采用先进的数据处理和分析技术。