| | |
Methods defined here:
- __init__(self, ConceptNetNLTools_handle=None, pred_filename='./generalised_predicates.txt')
- decode_node(self, node_tuple)
- decode_word(self, word_uid)
- display_node(self, textnode)
- returns the pretty print of a node's contents
- edgeuid2zipped(self, uid)
- encode_node(self, textnode)
- encode_word(self, word)
- get_all_projections(self, textnode_list)
- inputs a list of concepts
computes all available contextual projections
and returns a list of pairs, each of the form:
('ProjectionName',
(('concept1',score1), ('concept2,score2), ...)
)
- get_analogous_concepts(self, textnode, simple_results_p=0)
- -inputs a node
-uses structure-mapping to generate a list of
analogous concepts
-each analogous concept shares some structural features
with the input node
-the strength of an analogy is determined by the number
and weights of each feature. a weighting scheme is used
to disproportionately weight different relation types
and also weights a structural feature by the equation:
math.log(f+f2+0.5*(i+i2)+2,4), where f= # of utterances,
i = # of times inferred
- outputs a list of RESULTs rank-ordered by relevance
- each RESULT is a triple of the form:
('analogous concept', SHARED_STRUCTURES, SCORE)
- SCORE is a scalar valuation of quality of a result
(for now, this number does not have much external meaning)
- SHARED_STRUCTURES is a list of triples, each of the form:
('RelationType', 'target node', SCORE2)
- SCORE2 is a scalar valuation of the strength of a
particular shared structure
- if simple_results_p = 1, then output object is simply
a list of rank-ordered concepts
- get_context(self, textnode_list, max_node_visits=500, max_results=200, flow_pinch=300, linktype_weights_dict=None, textnode_list_weighted_p=0)
- the max_node_visits determines how far context will spread
increasing it adversely affects runtime
max_results limits the number of results returned
but changing it does not affect runtime
flow_pinch limits the number of edges considered
at each step of the context flow
the linktype_weights_dict is a python dictionary
whose keys are the conceptnet relationtypes and whose
values are a weight assigned to each, in the range [0.0,1.0]
- to blacklist a linktype, set its weight to 0.0
- context flow along backedges are regulated by entries in the
linktype_weights_dict whose key names are "relationtype"+"Inverse"
- considering inverse flows slow the runtime of this function a bit
- omitting a relationtype will default to it being blacklisted
- for reference, the default_linktype_weights_dict is:
default_linktype_weights_dict = {
'ConceptuallyRelatedTo':0.1,
'IsA':0.9,
'FirstSubeventOf':1.0,
'DesirousEffectOf':1.0,
'ThematicKLine':0.8,
'MadeOf':0.7,
'SubeventOf':0.9,
'UsedFor':1.0,
'SuperThematicKLine':1.0,
'DefinedAs':1.0,
'LastSubeventOf':1.0,
'LocationOf':0.9,
'CapableOfReceivingAction':0.6,
'CapableOf':0.8,
'PrerequisiteEventOf':1.0,
'MotivationOf':1.0,
'PropertyOf':1.0,
'PartOf':1.0,
'EffectOf':1.0,
'DesireOf':1.0,
'ConceptuallyRelatedToInverse':0.0,
'IsAInverse':0.0,
'FirstSubeventOfInverse':0.0,
'DesirousEffectOfInverse':0.0,
'ThematicKLineInverse':0.0,
'MadeOfInverse':0.0,
'SubeventOfInverse':0.0,
'UsedForInverse':0.0,
'SuperThematicKLineInverse':0.0,
'DefinedAsInverse':0.0,
'LastSubeventOfInverse':0.0,
'LocationOfInverse':0.0,
'CapableOfReceivingActionInverse':0.0,
'CapableOfInverse':0.0,
'PrerequisiteEventOfInverse':0.0,
'MotivationOfInverse':0.0,
'PropertyOfInverse':0.0,
'PartOfInverse':0.0,
'EffectOfInverse':0.0,
'DesireOfInverse':0.0,
}
if textnode_list_weighted_p, then each element of textnode_list
is not a string, but instead, of the form: ('dog',0.5)
where the cdr is the relative origin weight of that concept.
- getuid(self)
- load_predicates(self)
- nodeuid2zipped(self, uid)
- optimize_order(self)
- pp_predicate(self, pred)
- project_affective(self, textnode_list)
- -inputs a list of concepts
-computes the affective projection, which is
the emotional context and consequences underlying these concepts
-returns a rank-ordered list of concepts and their scores
e.g.: (('concept1',score1), ('concept2,score2), ...)
- project_consequences(self, textnode_list)
- -inputs a list of concepts
-computes the causal projection, which consists of
possible consequences of an event or possible actions
resulting from the presence of a thing
-returns a rank-ordered list of concepts and their scores
e.g.: (('concept1',score1), ('concept2,score2), ...)
- project_details(self, textnode_list)
- -inputs a list of concepts
-computes the detail projection, which consists of
a thing's parts, materials, properties, and instances
and an event's subevents
-returns a rank-ordered list of concepts and their scores
e.g.: (('concept1',score1), ('concept2,score2), ...)
- project_spatial(self, textnode_list)
- -inputs a list of concepts
-computes the spatial projection, which consists of
relevant locations, relevant objects in the same scene.
-returns a rank-ordered list of concepts and their scores
e.g.: (('concept1',score1), ('concept2,score2), ...)
- unpp(self, pp)
- zipped2edgeuid(self, edge)
- zipped2nodeuid(self, zipped)
|