| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document - typo error corrected |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| ---------------------------------------------------------------------------------------------------------------- ConceptNet 5.4 Python RESTful API implementation - lookup, search and association ---------------------------------------------------------------------------------------------------------------- (*) This commit implements the RESTful Python requests HTTP API for querying ConceptNet 5.4 dataset (*) Three functions for looking up a concept, searching a concept and finding similar associated concepts have been implemented with 3 endpoints (as per the documentation in https://github.com/commonsense/conceptnet5/wiki/API/2349f2bbd1d7fb726b3bbdc14cbeb18f0a40ef18) (*) ConceptNet is quite different from WordNet in representation as JSON dictionary as against graph in WordNet. (*) ConceptNet 5.4 endpoints have been invoked instead of 5. (*) If necessary ConceptNet can replace all WordNet invocations in a later point in time. But such a replacement is non-trivial and would probably require almost all Recursive Gloss Overlap related code to be rewrittern from scratch. But that depends on how ConceptNet 5 weighs against WordNet in measuring semantic meaningfulness. |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| ------------------------------------------------------------------------------------------------------------------------- Updates to Text Summarization - choosing sentences matching class labels - 6 July 2017 ------------------------------------------------------------------------------------------------------------------------- (*) Some experimentation on choosing relevant sentences to be added to summary was performed (*) relevance_to_text() invocation has been changed as per algorithm below: for each prominent dense subgraph k-core class label find the synset definition of class label for each sentence invoke relevance_to_text() similarity function between sentence and the class label definition and add to summary if the relevance ratio > 0.41 and if not already in summary (*) Ratio 0.41 was arrived at heuristically: - For relevance ratio 0.1, summary ratio was 0.65 - For relevance ratio 0.2, summary ratio was 0.48 - For relevance ratio 0.3, summary ratio was 0.35 - For relevance ratio 0.4, summary ratio was 0.21 - For relevance ratio 0.5, summary ratio was 0.02 There is a drastic dip in size of summary if relevance threshold is increased beyond 0.4. Because of this 0.41 has been hardcoded. (*) Logs for this have been added to testlogs/ (*) Summary generated is human readable - subset chosen from actual text. |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| ------------------------------------------------------------------------------------------------------------------------------------------ Updates to Text Summarization - Commits - 5 July 2017 ------------------------------------------------------------------------------------------------------------------------------------------ (*) Changed the class-sentence matching clause by increasing the percentile of prominent classes (*) Changed relevance_to_text() by comparing each sentence chosen by prominent class match to the sentences in text - Ratcliff-Obershelp Longest Common Subsequence matching difflib library function is invoked for this similarity (*) Percentage of summary to the size of the text has been printed as a ratio. Logs have been committed to testlogs/ |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| ---------------------------------------------------------------------------------------------------------------------- Updates to Text Summarization from Dense Subgraph of Recursive Gloss Overlap graph of a Text ---------------------------------------------------------------------------------------------------------------------- (*) import matplotlib commented in RGO classifier (*) New clause added to Text Summarization: This finds the common path between 2 word vertices synsets by inheriting an existing WordNet code for shortest wordnet path distance and constructs sentences by a sliding window for each successive pair of intermediate vertices in this common path. This creates a deeper profund summary than least_common_hypernyms().Sentences thus created are sorted based on a relevance score to the actual text - size of intersection similar to Lesk WSD. Only top relevant sentences can be chosen as required. (*) Logs for this have been committed to testlogs/. |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| ---------------------------------------------------------------------------------------------------------------------- Text Summarization from Recursive Gloss Overlap Graph Core ---------------------------------------------------------------------------------------------------------------------- (*) New clause added to classify the text and match the prominent core number word vertices in the text and only add those sentences to summary. (*) This is an alternative to k-core subgraph traversal and creating text programmatically. It is based on the heuristic that a summary should capture the essence/classes the text belongs to. (*) Prominent core number classes are shaved off from the sorted core number list returned by RGO classifier and top percentile is used. Summary is limited in size relative to the original text. |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| (*) Added a new function to create a summary from text - This function creates the core of a Recursive Gloss Overlap graph with certain core number and writes out a text sentence for each edge in core subgraph obtaining least common ancestor(hypernym) relation (*) logs for this have been committed to python-src/testlogs |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
| Updated AsFer Design Document |
|
More...
|
over 9 years ago
|
