def eval_avgs_against_descriptions(embedding=None):
"""
Compute ROUGE-L and cosine similarity evaluation measures where ideal
summaries are taken as the TED Talk descriptions.
:param embedding: fastText word embedding.
:return: List containing averages of precision, recall, F-score, and cosine
similarity over 50 documents.
:rtype: list(float)
"""
# Get records where one or more tag is present in transcript.
df = read_data(75)
df = drop_noconcept_recs(df)[:50]
results = []
for j in range(len(df)):
s = summarize(df.iloc[[j]], df['d_cnt'].iloc[[j]][j])
ideal = preprocess_transcripts(df.iloc[[j]],
clean_transcript_data=False,
df_field='description')
rl = Evaluation.rouge_l(s, ideal[0][0])
cs = Evaluation.cos_similarity(s, ideal[0][0])
results.append([rl, cs])
# Average evaluation scores over number of dataframe records.
results = np.asarray(results)
rlresults = results[:, 0]
cossim_results = results[:, 1]
avg_prec = np.average([rlresults[j][0] for j in range(results.shape[0])])
avg_recall = np.average([rlresults[j][1] for j in range(results.shape[0])])
avg_fscore = np.average([rlresults[j][2] for j in range(results.shape[0])])
avg_cossim = np.average(cossim_results)
return [avg_prec, avg_recall, avg_fscore, avg_cossim]
def eval_against_humangenerated(method, embedding=None):
"""
Compute ROUGE-L and cosine similarity evaluation measures for first five
records where ideal summaries are human generated.
:param method: LSA or TextRank summarization method.
:param embedding: fastText word embedding.
:return results: List containing evalution measure computations.
:rtype: list(array_type): float
"""
human_summaries = [
("It's never happened before in software! Remember, the "
"hard part is not deciding what features to add, it's "
"The lesson was: simplicity sells."),
("This is where I realized that there was really a need to communicate, "
"because the data of what's happening in the world and the child "
"health of every country is very well aware."
"Now, statisticians don't like it, because they say that this will not "
"show the reality; we have to have statistical, analytical methods. "
"And it's a new technology coming in, but then amazingly, how well it "
"fits to the economy of the countries."),
("And the interesting thing is: if you do it for love, the money comes "
"anyway. 'To be successful, put your nose down in something and get "
"damn good at it.' Persistence is the number one reason for our success."
),
("So honeybees are important for their role in the economy as well as "
"in agriculture. We need bees for the future of our cities and urban "
"living. What can you do to save the bees or to help them or to think "
"of sustainable cities in the future?"),
("So now I want to introduce you to my new hero in the global climate "
"change war, and that is the eastern oyster. So the oyster was the "
"basis for a manifesto-like urban design project that I did about the "
"New York Harbor called oyster-tecture. To conclude, this is just one "
"cross-section of one piece of city, but my dream is, my hope is, that "
"when you all go back to your own cities that we can start to work "
"together and collaborate on remaking and reforming a new urban "
"landscape towards a more sustainable, a more livable and a more "
"delicious future.")
]
df = read_data(5)
results = []
for j in range(len(df)):
s = method.summarize_text(df.iloc[[j]], 3)
rl = Evaluation.rouge_l(s, human_summaries[j])
cs = Evaluation.cos_similarity(s, human_summaries[j])
results.append([rl, cs])
return results