ITGumbo: spicing IT up

Consumer Sentiments

New Ontology

The new properties that may be required for expressing consumer sentiments in order to collect context data are shown in sample RDF triples:

<product IRI> std:affect "liked/disliked/good/bad/recommend/reject"
<product IRI> std:review "short comment"

Example

The blogger may use other terms such as recommend, suggest, etc. The object value for property std:affect may be any combination of verb, adverb and adjective. If the blog poster uses RDFa attribute about to identify the 'thing' discussed or referred in the post and attribute property to describe the 'thing' characteristics within the blog post then the data for consumer sentiments and context information can be gleaned from the blog post. Product IRI that is the subject of the RDF tripple shall be same in all blog posts, the predicate of this RDF tripple is std:affect.

Product Context Data

Reference of other products/services by bloggers in a blog post for a particular product/service can provide intelligent context data. E.g. mention of sunshine pearls in a blog post on hand bags provides context information on what size and color pearls match with a hand bag. A business opportunity for both handbags and pearls can be found by the analysis of this context data. The two products are bound to each other with the property dc:relation. The binding may be written in a reverse order as well, it depends on how the sentence is framed. The requirement is to provide semantic annotation of data and to provide marketing tools that can use this context data to find new business opportunity. The tool must be intelligent to resolve std:affect synonym values in order to produce context histograms.

SPARQL query

PREFIX std: <http://www.somestandards.org/>
PREFIX dc: <http://purl.org/dc/elements/1.1>
SELECT ?s ?o ?v
WHERE { 
	{ 
	  ?s dc:relation <pearl IRI> .
	} UNION
	{
	  <pearl IRI> dc:relation ?o .
	}
	<pearl IRI> ?p1 "small"@en .
	<pearl IRI> ?p2 "natural"@en .
	FILTER(?p1 = std:size && ?p2 = std:color)
	<pearl IRI> std:affect ?v .
      }

This SPARQL query shall select all web resources that are related to "pearl IRI" and have included the given properties for "pearl IRI". The SPARQL query solution shall also include the value of property std:affect used in the web resource (e.g. blog post). Blog category and tags also provide data that may be used to collect relevant information from this free source of online information. The properties that may be used for these are dcmitype:collection and dc:subject respectively.

Microformat vote-links and hreview may be embedded for rating. Vote-links microformats for similar purpose may be included as value of rev attribute. The hreview microformat 'rating' may also be used to provide numerical rating.

Conclusion: The advantage of subjective opinion on a product through std:affect property is that it has enhanced expressive power as compared to numerical value. The advantage of using RDFa over microformat are obviously the emerging semantic web applications. The purpose of the new properties is to enable annotation of language qualifiers within a sentence, the alternative may be to provide the value of the new properties as radio buttons. But the objective here is to capture the sentiments from within the text of the blog post. It is not possible to provide radio buttons for all products that may be referred to in blog posts. Without this semantic annotation of products and sentiments this useful context data may be lost.

Reference:
DCMI Metadata Terms, RDF, RDFa, SPARQL.
Leon, de. Ponce, Sandra. (2007) Sentiment Scoring Partnership With KD Paine. Buzzlogic.