Query related to incorrect output for get-neighbors for RandomIndexing

[rati...@gmail.com]

Hi,

I want to get the semantically similar neighbors to a term. I use RandomIndexing 

Command:

java -cp classes edu.ucla.sspace.mains.RandomIndexingMain -d <inputfile> --tokenFilter=exclude=english-stop-words-large.txt --verbose ~/Project/sspace/sspaceoutput/

the input corpus is English news text from http://www.statmt.org/wmt14/translation-task.html

It basically contains news headlines and other such text one document per line.

I train with around 2 million lines of text.

I do preprocessing on text to tokenize and wordnet stem the text.

For word "begin" I get the following output which makes sense

>get-neighbors begin

left 0.7131183273100001

die 0.7137912698526857

remain 0.7176777557552262

receive 0.7190962118578188

serve 0.7191618319236693

launch 0.71928316727257

complete 0.748564586407225

0.7565354741099785

turn 0.7669144379867078

start 0.8847017108603537

But for word "tiger"

> get-neighbors tiger

piney 0.5286385694809764

eighty-pound 0.5292671527978384

bisri 0.5298957361147005

flower-strewn 0.5305243194315625

templeton 0.6600891439647897

bretton 0.6678980395401771

twinberrow 0.7493857366570842

mid-tone 0.7493857366570842

106,200 0.7493857366570842

bashea 0.749385736657084

tiger occurs around 1000 times in the corpus

The word bashea only occurs once in the whole corpus. I am not able to understand why it ranked so highly 

Other outputs too are not very correlated.

Please help me in resolving the issue.

Thanks,

Ratish Puduppully