[READ-ONLY] Mirror of https://github.com/FoxxMD/string-sameness. Compare the sameness of two strings
foxxmd.github.io/string-sameness/
compare
cosine-similarity
dice-coefficient
levenshtein-distance
sameness
string
text
typescript
feat: Disable cosine for low token count strings
* Score is inaccurate when there are only a few words in the string so disable in order to keep combined strat score more accurate
* Add disclaimer to cosineSimilarity export docs
* Add matchingStrategies module to typedoc for visibility
+46
-6
+5
-3
src/index.ts
+5
-3
src/index.ts
···
1
-
import {cosineStrategy, levenStrategy, diceStrategy} from "./matchingStrategies/index.js";
1
+
import {cosineStrategy, levenStrategy, diceStrategy, cosineStrategyAggressive} from "./matchingStrategies/index.js";
2
2
import {
3
3
ComparisonStrategyResult,
4
4
NamedComparisonStrategyObjectResult,
···
84
84
const strategies = {
85
85
diceStrategy,
86
86
levenStrategy,
87
-
cosineStrategy
87
+
cosineStrategy,
88
+
cosineStrategyAggressive
88
89
};
89
90
90
91
export {
···
94
95
createStringSameness,
95
96
defaultStrategies,
96
97
strategies,
97
-
defaultStrCompareTransformFuncs
98
+
defaultStrCompareTransformFuncs,
99
+
ComparisonStrategyResult
98
100
}
+11
src/matchingStrategies/cosineSimilarity.d.ts
+11
src/matchingStrategies/cosineSimilarity.d.ts
···
1
1
import { ComparisonStrategy, ComparisonStrategyResultObject } from "../atomic.js";
2
2
export declare const calculateCosineSimilarity: (strA: string, strB: string) => number;
3
+
/**
4
+
* Compares whole tokens (words) within a string independent of order
5
+
*
6
+
* This strategy is automatically disabled for strings with less than 4 words because it can lead to inaccurate scores due to not comparing characters IE it is not very useful for short sentences and comparing single words with typos
7
+
*
8
+
* If you'd like to use it even in these scenarios build your own strategy array using cosineStrategyAggressive instead of this one
9
+
* */
3
10
export declare const cosineStrategy: ComparisonStrategy<ComparisonStrategyResultObject>;
11
+
/**
12
+
* Always runs (strings are always valid) which may lead to inaccurate scores in low token-count strings
13
+
* */
14
+
export declare const cosineStrategyAggressive: ComparisonStrategy<ComparisonStrategyResultObject>;
+27
-1
src/matchingStrategies/cosineSimilarity.ts
+27
-1
src/matchingStrategies/cosineSimilarity.ts
···
69
69
return cosineSimilarity(termFreqVecA, termFreqVecB);
70
70
}
71
71
72
-
export const cosineStrategy: ComparisonStrategy<ComparisonStrategyResultObject> = {
72
+
const cosineBaseStrategy: ComparisonStrategy<ComparisonStrategyResultObject> = {
73
73
name: 'cosine',
74
74
strategy: (valA: string, valB: string) => {
75
75
let res = calculateCosineSimilarity(valA, valB);
···
82
82
}
83
83
}
84
84
}
85
+
86
+
/**
87
+
* Compares whole tokens (words) within a string independent of order
88
+
*
89
+
* This strategy is automatically disabled for strings with less than 4 words because it can lead to inaccurate scores due to not comparing characters IE it is not very useful for short sentences and comparing single words with typos
90
+
*
91
+
* If you'd like to use it even in these scenarios build your own strategy array using cosineStrategyAggressive instead of this one
92
+
* */
93
+
export const cosineStrategy: ComparisonStrategy<ComparisonStrategyResultObject> = {
94
+
...cosineBaseStrategy,
95
+
isValid: (valA: string, valB: string) => {
96
+
// cosine only compares full tokens (words), rather than characters, in a string
97
+
// which makes its score very inaccurate when comparing low token-count strings (short sentences and/or words with typos)
98
+
// so disable its usage if there are less than 4 tokens
99
+
const valATokenLength = valA.split(' ').length;
100
+
const valBTokenLength = valB.split(' ').length;
101
+
return valATokenLength < 4 || valBTokenLength < 4;
102
+
}
103
+
}
104
+
105
+
/**
106
+
* Always runs (strings are always valid) which may lead to inaccurate scores in low token-count strings
107
+
* */
108
+
export const cosineStrategyAggressive: ComparisonStrategy<ComparisonStrategyResultObject> = {
109
+
...cosineBaseStrategy
110
+
}
+2
-1
src/matchingStrategies/index.ts
+2
-1
src/matchingStrategies/index.ts
···
1
1
import {levenStrategy} from "./levenSimilarity.js";
2
-
import {cosineStrategy} from "./cosineSimilarity.js";
2
+
import {cosineStrategy, cosineStrategyAggressive} from "./cosineSimilarity.js";
3
3
import {diceStrategy} from "./diceSimilarity.js";
4
4
import {ComparisonStrategy, ComparisonStrategyResultValue, ComparisonStrategyResultObject, StrategyFunc} from '../atomic.js';
5
5
6
6
export {
7
7
levenStrategy,
8
8
cosineStrategy,
9
+
cosineStrategyAggressive,
9
10
diceStrategy,
10
11
ComparisonStrategy,
11
12
ComparisonStrategyResultObject,