知识库文档处理，word转markdown

前一篇我讲解了如何对接MiniMax实现FAQ，其实知识库不仅仅可以实现FAQ，还能实现帮助文档的查询，内部培训资料的查询等等，但是这些培训资料大部分是word版本的，并且有层级结构，比如标题1，标题1-1等等，这种层级结构AI是无法识别的，所以需要转换成markdown格式的，帮助文档还有一个问题就是相似问题容易混淆，比如A文档有联系人，B文档也有联系人，这时查联系人就会出现错误的情况，所以要在每个标题前加前缀，比如A文档的联系人，B文档的联系人，这些也希望能够在转换程序中实现.

代码实现

public class MDTest {@Testpublic void testMD() throws Exception {String preHeader = "A文档的";if (preHeader == null) preHeader = "";String path = "d:/333.docx";File file = new File(path);FileInputStream is = new FileInputStream(file);XWPFDocument document = new XWPFDocument(is);StringBuilder sb = new StringBuilder();for (IBodyElement e : document.getBodyElements()) {if (e instanceof XWPFParagraph) {XWPFParagraph paragraph = (XWPFParagraph) e;appendParagraphText(sb, paragraph, document, preHeader);} else if (e instanceof XWPFTable) {XWPFTable table = (XWPFTable) e;appendTableText(sb, table);} else if (e instanceof XWPFSDT) {sb.append(((XWPFSDT) e).getContent().getText());}sb.append('\n');}FileUtil.writeBytes(sb.toString().getBytes(), "d:/aaa.md");}private static boolean isHeadTitle(XWPFDocument document, XWPFParagraph paragraph) {String styleName = getStyleName(document, paragraph);return styleName.toLowerCase().startsWith("heading");}private static String getStyleName(XWPFDocument document, XWPFParagraph paragraph) {String styleId = paragraph.getStyleID();if (StringUtils.isNotEmpty(styleId)) {XWPFStyle style = document.getStyles().getStyle(styleId);return style.getName();}return "";}private static void appendTableText(StringBuilder text, XWPFTable table) {for (XWPFTableRow row : table.getRows()) {List<ICell> cells = row.getTableICells();for (int i = 0; i < cells.size(); i++) {ICell cell = cells.get(i);if (cell instanceof XWPFTableCell) {text.append(((XWPFTableCell) cell).getTextRecursively());} else if (cell instanceof XWPFSDTCell) {text.append(((XWPFSDTCell) cell).getContent().getText());}if (i < cells.size() - 1) {text.append("\t");}}text.append('\n');}}private static void appendParagraphText(StringBuilder text, XWPFParagraph paragraph, XWPFDocument document, String preHeader) {String styleName = getStyleName(document, paragraph);if (styleName.toLowerCase().startsWith("heading")) {String number = styleName.replaceAll(".*(\\d+)$", "$1");appendHeader(text, number, preHeader);}for (IRunElement run : paragraph.getRuns()) {text.append(run);}}private static void appendHeader(StringBuilder text, String number, String preHeader) {if (StringUtils.isEmpty(number)) return;int num = Integer.parseInt(number);for (int i = 0; i < num; i++) {text.append("#");}text.append(" ").append(preHeader);}
}

代码解析

我们先通过poi读取word文档，然后获取段落，getStyleName方法可以获取段落的样式，如果样式是heading 1,表示是标题1，heading 2是标题2，而markdown格式的标题1是# 标题,标题2是## 标题，这样就能实现转换，其余的直接按照文本放入就行了

效果

word形式

markdown形式