use power mean to rescale weights

Project.toml

@@ -1,7 +1,7 @@
 name = "WordCloud"
 uuid = "6385f0a0-cb03-45b6-9089-4e0acc74b26b"
 authors = ["guoyongzhi <guo-yong-zhi@outlook.com>"]
-version = "0.10.11"
+version = "0.10.12"
 
 [deps]
 ColorSchemes = "35d6a980-a343-548e-a6ea-1d62b119f2f4"

plutoapp.jl

@@ -62,6 +62,9 @@ end
 # ╔═╡ b38c3ad9-7885-4af6-8394-877fde8ed83b
 md"**mask outline:** $(@bind outlinewidth NumberField(-1:100, default=-1))　*-1 means random*"
 
+# ╔═╡ 6e614caa-38dc-4028-b0a7-05f7030d5b43
+md"**layout style:** $(@bind style Select([:auto, :uniform, :gathering]))"
+
 # ╔═╡ 872f2653-303f-4b53-8e01-26bec86fc413
 md"""**text density:** $(@bind density NumberField(0.1:0.01:10.0, default=0.5))　　**min word spacing:** $(@bind spacing NumberField(0:100, default=2))"""
 
@@ -219,12 +222,11 @@ weightscale_funcs = [
     (n->n^2) => "x²",
     expm1 => "exp x",
     ]
-weightscale_types = [:wordarea, :fontsize, :diagonallength]
 nothing
 end
 
-# ╔═╡ 6e614caa-38dc-4028-b0a7-05f7030d5b43
-md"**layout style:** $(@bind style Select([:auto, :uniform, :gathering]))　　**rescale weights:** $(@bind rescale_func Select(weightscale_funcs)) $(@bind scalekeeptype Select(weightscale_types))"
+# ╔═╡ dfe608b0-077c-437a-adf2-b1382a0eb4eb
+md"**rescale weights:** $(@bind rescale_func Select(weightscale_funcs))　　**word length balance:** $(@bind word_length_balance Slider(-8:0.1:8, default=0))"
 
 # ╔═╡ e7ec8cd7-f60b-4eb0-88fc-76d694976f9d
 begin
@@ -309,7 +311,7 @@ try
 	else
 		text = read_table(text_)
 	end
-	dict_process = rescaleweights(rescale_func, scalekeeptype) ∘ casemerge! ∘ lemmatize!
+	dict_process = rescaleweights(rescale_func, word_length_balance) ∘ casemerge! ∘ lemmatize!
 	if text isa AbstractString && ischinese(text)
 		println("检测到中文")
 		text = wordseg_cn(text)
@@ -421,6 +423,7 @@ end
 # ╟─1aa632dc-b3e8-4a9d-9b9e-c13cd05cf97e
 # ╟─b38c3ad9-7885-4af6-8394-877fde8ed83b
 # ╟─6e614caa-38dc-4028-b0a7-05f7030d5b43
+# ╟─dfe608b0-077c-437a-adf2-b1382a0eb4eb
 # ╟─872f2653-303f-4b53-8e01-26bec86fc413
 # ╟─26d6b795-1cc3-4548-aa07-86c2f6ee0776
 # ╟─7993fd44-2fcf-488e-9280-4b4d0bf0e22c

src/textprocessing.jl

@@ -104,15 +104,16 @@ function lemmatize!(d::AbstractDict)
     d
 end
 
-function _rescaleweights(dict, func=identity, keep=:wordarea)
-    @assert keep in [:wordarea, :fontsize, :diagonallength]
-    if keep == :wordarea
-        newdict = Dict(k => func(v) / sqrt(length(k)) for (k, v) in dict)
-    elseif keep == :fontsize
-        newdict = Dict(k => func(v) for (k, v) in dict)
+function _rescaleweights(dict, func=identity, p=0) # p is the exponent of the power mean
+    # ((fontsize^p + (wordlength*fontsize)^p)/2) ^ (1/p) = weight
+    # p=-1, harmonic mean; p=0, geometric mean; p=1, arithmetic mean; p=2, root mean square;
+    # p=-∞, minimum; p=∞, maximum;
+    if p == 0
+        wordlength_scale = l->sqrt(l)
     else
-        newdict = Dict(k => func(v) / sqrt(length(k)^2 + 1) for (k, v) in dict)
+        wordlength_scale = l->(l^p+1)^(1/p)
     end
+    newdict = Dict(k => func(v) / wordlength_scale(length(k)) for (k, v) in dict)
     sc = sum(values(dict)) / sum(values(newdict))
     for k in keys(newdict)
         newdict[k] *= sc
@@ -140,7 +141,7 @@ function processtext(counter::AbstractDict{<:AbstractString,<:Real};
     minfrequency=0,
     maxnum=500,
     minweight=1 / maxnum, maxweight=:auto,
-    process=rescaleweights(identity, :wordarea) ∘ casemerge! ∘ lemmatize!)
+    process=rescaleweights(identity, p=0) ∘ casemerge! ∘ lemmatize!)
     stopwords isa AbstractSet || (stopwords = Set(stopwords))
     counter = process(counter)
     print("Total words: $(round(sum(values(counter)), digits=2)). ")