This repository was archived by the owner on Aug 13, 2026. It is now read-only.
-
-
Notifications
You must be signed in to change notification settings - Fork 25k
Expand file tree
/
Copy pathUnstructured.ts
More file actions
176 lines (146 loc) · 6.08 KB
/
Copy pathUnstructured.ts
File metadata and controls
176 lines (146 loc) · 6.08 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import {
HiResModelName,
SkipInferTableTypes,
UnstructuredLoaderOptions,
UnstructuredLoaderStrategy
} from '@langchain/community/document_loaders/fs/unstructured'
import { BaseDocumentLoader } from 'langchain/document_loaders/base'
import { StringWithAutocomplete } from 'langchain/dist/util/types'
import { Document } from '@langchain/core/documents'
/**
* Set the chunking_strategy to chunk text into larger or smaller elements. Defaults to None with optional arg of by_title
*/
type ChunkingStrategy = 'None' | 'by_title'
/**
* Represents an element returned by the Unstructured API. It has
* properties for the element type, text content, and metadata.
*/
type Element = {
type: string
text: string
// this is purposefully loosely typed
metadata: {
[key: string]: unknown
}
}
export class UnstructuredLoader extends BaseDocumentLoader {
public filePath: string
private apiUrl = process.env.UNSTRUCTURED_API_URL || 'https://api.unstructuredapp.io/general/v0/general'
private apiKey: string | undefined = process.env.UNSTRUCTURED_API_KEY
private strategy: StringWithAutocomplete<UnstructuredLoaderStrategy> = 'hi_res'
private encoding?: string
private ocrLanguages: Array<string> = []
private coordinates?: boolean
private pdfInferTableStructure?: boolean
private xmlKeepTags?: boolean
private skipInferTableTypes?: Array<StringWithAutocomplete<SkipInferTableTypes>>
private hiResModelName?: StringWithAutocomplete<HiResModelName>
private includePageBreaks?: boolean
private chunkingStrategy?: StringWithAutocomplete<ChunkingStrategy>
private multiPageSections?: boolean
private combineUnderNChars?: number
private newAfterNChars?: number
private maxCharacters?: number
constructor(optionsOrLegacyFilePath: UnstructuredLoaderOptions) {
super()
const options = optionsOrLegacyFilePath
this.apiKey = options.apiKey
this.apiUrl = options.apiUrl || this.apiUrl
this.strategy = options.strategy || this.strategy
this.encoding = options.encoding
this.ocrLanguages = options.ocrLanguages || this.ocrLanguages
this.coordinates = options.coordinates
this.pdfInferTableStructure = options.pdfInferTableStructure
this.xmlKeepTags = options.xmlKeepTags
this.skipInferTableTypes = options.skipInferTableTypes
this.hiResModelName = options.hiResModelName
this.includePageBreaks = options.includePageBreaks
this.chunkingStrategy = options.chunkingStrategy
this.multiPageSections = options.multiPageSections
this.combineUnderNChars = options.combineUnderNChars
this.newAfterNChars = options.newAfterNChars
this.maxCharacters = options.maxCharacters
}
async _partition(buffer: Buffer, fileName: string): Promise<Element[]> {
const formData = new FormData()
formData.append('files', new Blob([buffer]), fileName)
formData.append('strategy', this.strategy)
this.ocrLanguages.forEach((language) => {
formData.append('ocr_languages', language)
})
if (this.encoding) {
formData.append('encoding', this.encoding)
}
if (this.coordinates === true) {
formData.append('coordinates', 'true')
}
if (this.pdfInferTableStructure === true) {
formData.append('pdf_infer_table_structure', 'true')
}
if (this.xmlKeepTags === true) {
formData.append('xml_keep_tags', 'true')
}
if (this.skipInferTableTypes) {
formData.append('skip_infer_table_types', JSON.stringify(this.skipInferTableTypes))
}
if (this.hiResModelName) {
formData.append('hi_res_model_name', this.hiResModelName)
}
if (this.includePageBreaks) {
formData.append('include_page_breaks', 'true')
}
if (this.chunkingStrategy) {
formData.append('chunking_strategy', this.chunkingStrategy)
}
if (this.multiPageSections !== undefined) {
formData.append('multipage_sections', this.multiPageSections ? 'true' : 'false')
}
if (this.combineUnderNChars !== undefined) {
formData.append('combine_under_n_chars', String(this.combineUnderNChars))
}
if (this.newAfterNChars !== undefined) {
formData.append('new_after_n_chars', String(this.newAfterNChars))
}
if (this.maxCharacters !== undefined) {
formData.append('max_characters', String(this.maxCharacters))
}
const headers = {
'UNSTRUCTURED-API-KEY': this.apiKey || ''
}
const response = await fetch(this.apiUrl, {
method: 'POST',
body: formData,
headers
})
if (!response.ok) {
throw new Error(`Failed to partition file ${this.filePath} with error ${response.status} and message ${await response.text()}`)
}
const elements = await response.json()
if (!Array.isArray(elements)) {
throw new Error(`Expected partitioning request to return an array, but got ${elements}`)
}
return elements.filter((el) => typeof el.text === 'string') as Element[]
}
async loadAndSplitBuffer(buffer: Buffer, fileName: string): Promise<Document[]> {
const elements = await this._partition(buffer, fileName)
const documents: Document[] = []
for (const element of elements) {
const { metadata, text } = element
if (typeof text === 'string') {
documents.push(
new Document({
pageContent: text,
metadata: {
...metadata,
category: element.type
}
})
)
}
}
return documents
}
async load(): Promise<Document[]> {
return Promise.reject(new Error('load() is not supported for UnstructuredLoader. Use loadAndSplitBuffer() instead.'))
}
}