Merge pull request #313 from kreuzberg-dev/fix-issue-309

fix(elixir): resolve DOCX keyword extraction FunctionClauseError

Author: Goldziher

CHANGELOG.md

@@ -93,6 +93,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Total 32 public types now properly exported for IDE autocomplete and type checking
   - Resolves import failures where types were defined but not accessible
 
+#### Elixir
+- **DOCX keyword extraction**: Fixed `FunctionClauseError` when extracting DOCX files with keywords metadata ([#309](https://github.com/kreuzberg-dev/kreuzberg/issues/309))
+  - DOCX extractor now parses comma-separated keyword strings into `Vec<String>` and stores in typed `Metadata.keywords` field
+  - Added defensive string handling to `normalize_keywords/1` in Elixir binding
+  - Resolves crash when extracting DOCX files containing keywords in `cp:keywords` or `dc:subject` metadata fields
+  - Added comprehensive unit tests for keyword string parsing in both Rust and Elixir
+
 ---
 
 ## [4.0.8] - 2026-01-17

crates/kreuzberg/src/extractors/docx.rs

@@ -182,6 +182,7 @@ impl DocumentExtractor for DocxExtractor {
         };
 
         let mut metadata_map = std::collections::HashMap::new();
+        let mut parsed_keywords: Option<Vec<String>> = None;
 
         if let Ok(core) = office_metadata::extract_core_properties(&mut archive) {
             if let Some(title) = core.title {
@@ -198,7 +199,14 @@ impl DocumentExtractor for DocxExtractor {
                 metadata_map.insert("subject".to_string(), serde_json::Value::String(subject));
             }
             if let Some(keywords) = core.keywords {
-                metadata_map.insert("keywords".to_string(), serde_json::Value::String(keywords));
+                // Parse comma-separated keywords into Vec<String>
+                parsed_keywords = Some(
+                    keywords
+                        .split(',')
+                        .map(|s| s.trim().to_string())
+                        .filter(|s| !s.is_empty())
+                        .collect(),
+                );
             }
             if let Some(description) = core.description {
                 metadata_map.insert("description".to_string(), serde_json::Value::String(description));
@@ -296,6 +304,7 @@ impl DocumentExtractor for DocxExtractor {
             mime_type: mime_type.to_string(),
             metadata: Metadata {
                 pages: page_structure,
+                keywords: parsed_keywords,
                 additional: metadata_map,
                 ..Default::default()
             },

crates/kreuzberg/tests/docx_metadata_extraction_test.rs

@@ -120,3 +120,133 @@ async fn test_docx_minimal_metadata_extraction() {
 
     println!("✅ DOCX minimal metadata extraction test passed!");
 }
+
+#[tokio::test]
+async fn test_docx_keywords_extraction() {
+    // This test verifies that DOCX keywords metadata is properly parsed
+    // from comma-separated strings into Vec<String> in Metadata.keywords
+    //
+    // Addresses GitHub issue #309: DOCX keyword extraction was returning
+    // strings instead of parsed keyword lists, causing FunctionClauseError
+    // in the Elixir binding.
+
+    use std::io::Write;
+    use tempfile::NamedTempFile;
+    use zip::CompressionMethod;
+    use zip::write::{FileOptions, ZipWriter};
+
+    // Create a minimal DOCX with keywords metadata
+    let mut temp_file = NamedTempFile::new().expect("Failed to create temp file");
+
+    {
+        let mut zip = ZipWriter::new(&mut temp_file);
+        let options: FileOptions<()> = FileOptions::default().compression_method(CompressionMethod::Stored);
+
+        // Add [Content_Types].xml
+        zip.start_file("[Content_Types].xml", options).unwrap();
+        zip.write_all(br#"<?xml version="1.0" encoding="UTF-8"?>
+<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types">
+  <Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/>
+  <Default Extension="xml" ContentType="application/xml"/>
+  <Override PartName="/word/document.xml" ContentType="application/vnd.openxmlformats-officedocument.wordprocessingml.document.main+xml"/>
+  <Override PartName="/docProps/core.xml" ContentType="application/vnd.openxmlformats-package.core-properties+xml"/>
+</Types>"#).unwrap();
+
+        // Add _rels/.rels
+        zip.start_file("_rels/.rels", options).unwrap();
+        zip.write_all(br#"<?xml version="1.0" encoding="UTF-8"?>
+<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">
+  <Relationship Id="rId1" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="word/document.xml"/>
+  <Relationship Id="rId2" Type="http://schemas.openxmlformats.org/package/2006/relationships/metadata/core-properties" Target="docProps/core.xml"/>
+</Relationships>"#).unwrap();
+
+        // Add word/document.xml with simple content
+        zip.start_file("word/document.xml", options).unwrap();
+        zip.write_all(
+            br#"<?xml version="1.0" encoding="UTF-8"?>
+<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
+  <w:body>
+    <w:p>
+      <w:r>
+        <w:t>Test document for keyword extraction</w:t>
+      </w:r>
+    </w:p>
+  </w:body>
+</w:document>"#,
+        )
+        .unwrap();
+
+        // Add docProps/core.xml with keywords (comma-separated string)
+        zip.start_file("docProps/core.xml", options).unwrap();
+        zip.write_all(
+            br#"<?xml version="1.0" encoding="UTF-8"?>
+<cp:coreProperties xmlns:cp="http://schemas.openxmlformats.org/package/2006/metadata/core-properties"
+                   xmlns:dc="http://purl.org/dc/elements/1.1/"
+                   xmlns:dcterms="http://purl.org/dc/terms/">
+  <dc:title>Test Document</dc:title>
+  <dc:creator>Test Author</dc:creator>
+  <cp:keywords>rust, docx, extraction, metadata, test</cp:keywords>
+  <dc:subject>Testing keyword extraction</dc:subject>
+</cp:coreProperties>"#,
+        )
+        .unwrap();
+
+        zip.finish().unwrap();
+    }
+
+    // Extract the DOCX file
+    let result = extract_file(
+        temp_file.path(),
+        Some("application/vnd.openxmlformats-officedocument.wordprocessingml.document"),
+        &ExtractionConfig::default(),
+    )
+    .await
+    .expect("Should extract DOCX with keywords successfully");
+
+    // Verify content was extracted
+    assert!(!result.content.is_empty(), "Content should not be empty");
+    assert!(
+        result.content.contains("Test document for keyword extraction"),
+        "Content should match document text"
+    );
+
+    // Verify keywords were parsed into Vec<String> in Metadata.keywords
+    assert!(
+        result.metadata.keywords.is_some(),
+        "Keywords should be present in metadata.keywords"
+    );
+
+    let keywords = result.metadata.keywords.as_ref().unwrap();
+    assert_eq!(
+        keywords.len(),
+        5,
+        "Should have 5 keywords parsed from comma-separated string"
+    );
+
+    // Verify individual keywords were trimmed and parsed correctly
+    assert_eq!(keywords[0], "rust", "First keyword should be 'rust'");
+    assert_eq!(keywords[1], "docx", "Second keyword should be 'docx'");
+    assert_eq!(keywords[2], "extraction", "Third keyword should be 'extraction'");
+    assert_eq!(keywords[3], "metadata", "Fourth keyword should be 'metadata'");
+    assert_eq!(keywords[4], "test", "Fifth keyword should be 'test'");
+
+    // Verify other metadata was also extracted
+    assert_eq!(
+        result.metadata.additional.get("created_by").and_then(|v| v.as_str()),
+        Some("Test Author"),
+        "Should have correct creator"
+    );
+    assert_eq!(
+        result.metadata.additional.get("title").and_then(|v| v.as_str()),
+        Some("Test Document"),
+        "Should have correct title"
+    );
+    assert_eq!(
+        result.metadata.additional.get("subject").and_then(|v| v.as_str()),
+        Some("Testing keyword extraction"),
+        "Should have correct subject"
+    );
+
+    println!("✅ DOCX keywords extraction test passed!");
+    println!("   Extracted keywords: {:?}", keywords);
+}

packages/elixir/lib/kreuzberg/result.ex

@@ -262,11 +262,19 @@ defmodule Kreuzberg.ExtractionResult do
   end
 
   @doc false
-  @spec normalize_keywords(list() | nil) :: list(map()) | nil
+  @spec normalize_keywords(list() | String.t() | nil) :: list(map()) | nil
   defp normalize_keywords(nil), do: nil
   defp normalize_keywords([]), do: []
   defp normalize_keywords(keywords) when is_list(keywords), do: keywords
 
+  defp normalize_keywords(keywords) when is_binary(keywords) do
+    keywords
+    |> String.split(",")
+    |> Enum.map(&String.trim/1)
+    |> Enum.reject(&(&1 == ""))
+    |> Enum.map(fn text -> %{"text" => text, "score" => 1.0} end)
+  end
+
   @doc false
   @spec normalize_elements(list() | nil) :: list(Kreuzberg.Element.t()) | nil
   defp normalize_elements(nil), do: nil

packages/elixir/test/unit/extraction_result_test.exs

@@ -407,4 +407,91 @@ defmodule KreuzbergTest.Unit.ExtractionResultTest do
       assert Enum.uniq(contents) == contents
     end
   end
+
+  describe "normalize_keywords/1 - keyword string parsing (GitHub issue #309)" do
+    test "parses comma-separated keyword string from DOCX metadata" do
+      # This test addresses GitHub issue #309: DOCX files return keywords
+      # as comma-separated strings from metadata, which caused FunctionClauseError
+      # before the fix was implemented.
+      keywords_string = "calibre, docs, ebook, conversion"
+
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: keywords_string)
+
+      assert is_list(result.keywords)
+      assert length(result.keywords) == 4
+
+      assert Enum.at(result.keywords, 0) == %{"text" => "calibre", "score" => 1.0}
+      assert Enum.at(result.keywords, 1) == %{"text" => "docs", "score" => 1.0}
+      assert Enum.at(result.keywords, 2) == %{"text" => "ebook", "score" => 1.0}
+      assert Enum.at(result.keywords, 3) == %{"text" => "conversion", "score" => 1.0}
+    end
+
+    test "parses keyword string with extra whitespace" do
+      keywords_string = "  keyword1  ,   keyword2   , keyword3  "
+
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: keywords_string)
+
+      assert is_list(result.keywords)
+      assert length(result.keywords) == 3
+
+      # Verify whitespace is properly trimmed
+      assert Enum.at(result.keywords, 0) == %{"text" => "keyword1", "score" => 1.0}
+      assert Enum.at(result.keywords, 1) == %{"text" => "keyword2", "score" => 1.0}
+      assert Enum.at(result.keywords, 2) == %{"text" => "keyword3", "score" => 1.0}
+    end
+
+    test "handles keyword string with trailing/leading commas" do
+      keywords_string = ",keyword1,keyword2,"
+
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: keywords_string)
+
+      assert is_list(result.keywords)
+      # Empty strings from leading/trailing commas should be filtered out
+      assert length(result.keywords) == 2
+      assert Enum.at(result.keywords, 0) == %{"text" => "keyword1", "score" => 1.0}
+      assert Enum.at(result.keywords, 1) == %{"text" => "keyword2", "score" => 1.0}
+    end
+
+    test "handles empty keyword string" do
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: "")
+
+      assert result.keywords == []
+    end
+
+    test "handles keyword string with only whitespace" do
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: "   ")
+
+      assert result.keywords == []
+    end
+
+    test "handles single keyword in string" do
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: "single")
+
+      assert is_list(result.keywords)
+      assert length(result.keywords) == 1
+      assert Enum.at(result.keywords, 0) == %{"text" => "single", "score" => 1.0}
+    end
+
+    test "assigns default score of 1.0 to parsed keywords" do
+      # Keywords from DOCX metadata don't have scores, so we assign default 1.0
+      keywords_string = "keyword1, keyword2"
+
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: keywords_string)
+
+      Enum.each(result.keywords, fn keyword ->
+        assert keyword["score"] == 1.0
+      end)
+    end
+
+    test "preserves keyword order from string" do
+      keywords_string = "first, second, third, fourth"
+
+      result = ExtractionResult.new("content", "text/plain", %{}, [], keywords: keywords_string)
+
+      assert Enum.at(result.keywords, 0)["text"] == "first"
+      assert Enum.at(result.keywords, 1)["text"] == "second"
+      assert Enum.at(result.keywords, 2)["text"] == "third"
+      assert Enum.at(result.keywords, 3)["text"] == "fourth"
+    end
+  end
 end