Python开发需求：为候选单词匹配种子词并实现打分功能

Alright, let's tackle this problem head-on. From what you've shared, you need to pair each word in your candidates list with relevant entries from seeds and calculate a score for each pair. I'll cover two practical, common approaches—one based on co-occurrence in your tweet corpus (super straightforward for your given data) and another based on semantic similarity (more flexible if you need broader relationships).

This method counts how many times a candidate word and a seed word appear together in the same tweet—perfect if your score should reflect how often they're mentioned alongside each other in your data.

Step-by-Step Implementation (Python) #

First, we'll normalize the text to avoid case sensitivity, then loop through each candidate-seed pair to tally co-occurrences:

candidates = ["you", "the", "best", "love", "fun", "feeling", "emotionally"]
seeds = ["happy", "love", "enjoy", "fun", "grace", "sad", "guilty"]
tweets = ["you look so happy", "I am in love with you", "hey you do the best at having fun okay", "i am emotionally sad right now", "feeling guilty"]

# Normalize all tweets to lowercase for consistent matching
lower_tweets = [tweet.lower() for tweet in tweets]

# Calculate co-occurrence scores
cooccurrence_scores = {}
for candidate in candidates:
    cooccurrence_scores[candidate] = {}
    for seed in seeds:
        count = 0
        for tweet in lower_tweets:
            # Check if both words are present in the tweet
            if candidate in tweet.split() and seed in tweet.split():
                count += 1
        cooccurrence_scores[candidate][seed] = count

# Print the results
for candidate, scores in cooccurrence_scores.items():
    print(f"Candidate: {candidate}")
    for seed, score in scores.items():
        print(f"  ↳ Seed '{seed}': {score}")

Sample Output Snippet #

Candidate: you
  ↳ Seed 'happy': 1
  ↳ Seed 'love': 1
  ↳ Seed 'enjoy': 0
  ...
Candidate: love
  ↳ Seed 'happy': 0
  ↳ Seed 'love': 1
  ...

If you want scores that reflect how similar the words are in meaning (not just how often they appear together), use pre-trained word embeddings. This works even if the pair never shows up in your tweets.

Step-by-Step Implementation (Python with spaCy) #

We'll use spaCy's pre-trained medium English model, which includes word vectors for semantic similarity:

import spacy

# Load the pre-trained model (run `pip install spacy` and `python -m spacy download en_core_web_md` first)
nlp = spacy.load("en_core_web_md")

candidates = ["you", "the", "best", "love", "fun", "feeling", "emotionally"]
seeds = ["happy", "love", "enjoy", "fun", "grace", "sad", "guilty"]

# Calculate semantic similarity scores
similarity_scores = {}
for candidate in candidates:
    similarity_scores[candidate] = {}
    candidate_doc = nlp(candidate)
    for seed in seeds:
        seed_doc = nlp(seed)
        # Cosine similarity (ranges from 0 to 1, higher = more similar)
        similarity_scores[candidate][seed] = round(candidate_doc.similarity(seed_doc), 3)

# Print the results
for candidate, scores in similarity_scores.items():
    print(f"Candidate: {candidate}")
    for seed, score in scores.items():
        print(f"  ↳ Seed '{seed}': {score}")

Sample Output Snippet #

Candidate: love
  ↳ Seed 'happy': 0.535
  ↳ Seed 'love': 1.0
  ↳ Seed 'enjoy': 0.618
  ...
Candidate: emotionally
  ↳ Seed 'sad': 0.387
  ↳ Seed 'guilty': 0.291
  ...

Which Approach Should You Use? #

• Go with co-occurrence if your score needs to be tied directly to the context of your provided tweets.
• Use semantic similarity if you want to capture broader linguistic relationships, even for word pairs that don't appear together in your data.

内容的提问来源于stack exchange，提问作者Indra