Solr ClassificationUpdateProcessorFactory Bayes: problem with labels - solr

Question

I have encountered a very strange behaviour: to test the classifcation function in Solr, I have defined the following processor chain:
<updateRequestProcessorChain name="classification">
<processor class="solr.UUIDUpdateProcessorFactory">
<str name="fieldName">id</str>
</processor>
<processor class="solr.ClassificationUpdateProcessorFactory">
<str name="inputFields">content</str>
<str name="classField">cat_knn</str>
<str name="predictedClassField.maxCount">2</str>
<str name="algorithm">knn</str>
<str name="knn.k">10</str>
<str name="knn.minTf">1</str>
<str name="knn.minDf">1</str>
</processor>
<processor class="solr.ClassificationUpdateProcessorFactory">
<str name="inputFields">content</str>
<str name="classField">cat_bayes</str>
<str name="predictedClassField.maxCount">2</str>
<str name="algorithm">bayes</str>
</processor>
<processor class="solr.LogUpdateProcessorFactory"/>
<processor class="solr.RunUpdateProcessorFactory"/>
</updateRequestProcessorChain>
As a test set I am using news categories, such as "business", "entertainment" etc.
The relevant fields are defined as follows:
<field name="cat_knn" type="text_en" indexed="true" stored="true" multiValued="true"/>
<field name="cat_bayes" type="text_en" indexed="true" stored="true" multiValued="true"/>
For the training set cat_knn and cat_bayes contain exactly the same category labels.
However, if I use the above chain to classify new documents, the cat_knn for new documents are labeled with the full label, i.e. "business" or "entertainment", whereas for the bayes algorithm the labels are cut and displayed as "busi" or "entertain". At the same time, a label like "sport" is properly recorded as "sport".
Any idea what might be going on here?

Answer 1

What you are seeing is the stemmed tokens for the field instead. On the SolrClassification wiki page it specifies that:
The field that contains the class of the document. It must appear in the indexed documents. If knn algorithm it must be stored. If bayes algorithm it must be indexed and ideally not heavily analysed.
This indicates that bayes uses the actual tokens, while knn uses the stored text for the field when outputting the class.
Change the field type to string or strings (single valued vs multivalued), or a text field with minimal analysis (maybe a KeywordTokenizer with only a LowercaseFilter or similar).