add a cosine similarity analysis cos_similarity()

JuliaText · Jan 6, 2021 · 7b3ce8c · 7b3ce8c
1 parent 992596b
commit 7b3ce8c
Show file tree

Hide file tree

Showing 3 changed files with 53 additions and 10 deletions.
diff --git a/src/TextAnalysis.jl b/src/TextAnalysis.jl
@@ -48,7 +48,7 @@ module TextAnalysis
     export hash_dtv, each_hash_dtv, hash_dtm, hash_tdm
     export CooMatrix, coom
     export standardize!
-    export tf, tf_idf, bm_25, lsa, lda, summarize
+    export tf, tf_idf, bm_25, lsa, lda, summarize, cos_similarity
     export tf!, tf_idf!, bm_25!, lda!
     export remove_patterns!, remove_patterns
     export prune!
@@ -61,7 +61,7 @@ module TextAnalysis
     export tag_scheme!
     export rouge_l_summary, rouge_l_sentence, rouge_n
     export PerceptronTagger, fit!, predict
-    
+
     export Vocabulary, lookup, update
     export everygram, padding_ngram
     export maskedscore, logscore, entropy, perplexity
@@ -93,14 +93,14 @@ module TextAnalysis
     include("coom.jl")
 
 
-    
+
     # Lang_model
     include("LM/vocab.jl")
-    include("LM/langmodel.jl") 
+    include("LM/langmodel.jl")
     include("LM/api.jl")
     include("LM/counter.jl")
     include("LM/preprocessing.jl")
-    
+
 
 
     function __init__()

diff --git a/src/tf_idf.jl b/src/tf_idf.jl
@@ -301,3 +301,33 @@ function tf_bm25!(dtm::AbstractMatrix{T}, tf::AbstractMatrix{F}
     end
     return tf
 end
+
+"""
+    function cos_similarity(tfidf::AbstractMatrix)
+
+`cos_similarity` calculates the cosine similarity from a term matrix (typically the tf-idf matrix).
+
+# Example
+```
+crps = Corpus( StringDocument.([
+    "to be or not to be",
+    "to sing or not to sing",
+    "to talk or to silence"]) )
+update_lexicon!(crps)
+d = dtm(crps)
+tfm = tf_idf(d)
+cs = cos_similarity(tfm)
+Matrix(cs)
+    # 3×3 Array{Float64,2}:
+    #  1.0        0.0329318  0.0
+    #  0.0329318  1.0        0.0
+    #  0.0        0.0        1.0
+```
+"""
+function cos_similarity(tfm::AbstractMatrix)
+    cs = tfm * tfm'
+    d = sqrt.(diag(cs))
+    # prevent division by zero  (only occurs for empty documents)
+    d[findall(iszero, d)] .= 1
+    cs ./ (d * d')
+end
diff --git a/test/tf_idf.jl b/test/tf_idf.jl
@@ -6,10 +6,7 @@
     doc3 = ""
     doc4 = "another another text text text text"
 
-    # TODO: this should work!
-    # crps = Corpus(map(StringDocument, [doc1 doc2 doc3 doc4]))
-
-    crps = Corpus(Any[StringDocument(doc1), StringDocument(doc2), StringDocument(doc3), StringDocument(doc4)])
+    crps = Corpus(StringDocument.([doc1, doc2, doc3, doc4]))
 
     update_lexicon!(crps)
     m = DocumentTermMatrix(crps)
@@ -76,7 +73,7 @@
         doc3 = ""
         doc4 = "another another text text text text"
 
-        crps = Corpus(Any[StringDocument(doc1), StringDocument(doc2), StringDocument(doc3), StringDocument(doc4)])
+        crps = Corpus(StringDocument.([doc1, doc2, doc3, doc4]))
 
         update_lexicon!(crps)
         m = DocumentTermMatrix(crps)
@@ -120,4 +117,20 @@
 
         @test_throws MethodError bm_25!(DocumentTermMatrix(crps))
     end
+
+	@testset "cosine similarity `cos_similarity()`" begin
+		crps = Corpus( StringDocument.([
+		    "to be or not to be",
+		    "to sing or not to sing",
+		    "to talk or to silence"]) )
+		update_lexicon!(crps)
+		d = dtm(crps)
+		tfm = tf_idf(d)
+		cs = cos_similarity(tfm)
+		@test cs==[
+			1.0                  0.03293177886245518  0.0;
+			0.03293177886245518  1.0                  0.0;
+			0.0                  0.0                  1.0
+		]
+    end
 end