You are provided with a paragraph. Your goal is to automatically identify the…
- User
System: You are provided with a paragraph. Your goal is to automatically identify the most relevant tasks based on the content of the paragraph and generate an example of entailment (if the task conclusion can be logically inferred from the paragraph) and a different example of non-entailment (if it is unclear or the opposite conclusion is true).
Step 1: Read the paragraph carefully. Step 2: Automatically decide which tasks are relevant based on the paragraph’s content. Relevant tasks may include, but are not limited to: - Topic Classification - Risk Classification - Sentiment Classification - Tone Classification - Event Detection - Entity Recognition - Geographical Recognition - Trend Detection - Intent Detection - Sector Identification - Impact Classification - Tense Identification (Past, Present, Future) - Uncertainty Identification
Step 3: For each relevant task you identify, generate two clear and concise examples whose labels match common practices for the task at hand: one for entailment and one for non-entailment. The example should speak about the text at the 3rd person. Step 4: Output the results in the following JSON format:
{ 'relevant_tasks': [ { 'task': , 'examples': [ { 'input': , 'label': 'entailment', }, { 'input': <The non-entailment example (neutral or contradiction) based on the paragraph>, 'label': 'non-entailment', } ] }, ... ] } User: Bristol Myers Squibb reported quarterly adjusted earnings that topped expectations and posted revenue in line with estimates. But sales of the company's popular blood cancer drug Revlimid plummeted due to generic competition. Bristol Myers Squibb on Thursday reported quarterly adjusted earnings Assistant:
- Assistant
{ "relevant_tasks": [ { "task": "Topic Classification", "examples": [ { "input": "The paragraph is about earnings reports and pharmaceutical revenue.", "label": "entailment" }, { "input": "The paragraph is about climate change policies.", "label": "non-entailment" } ] }, { "task": "Sentiment Classification", "examples": [ { "input": "The sentiment of the paragraph is neutral, as it reports on both positive and negative financial outcomes.", "label": "entailment" }, { "input": "The sentiment of the paragraph is overwhelmingly positive with no negative aspects.", "label": "non-entailment" } ] }, { "task": "Event Detection", "examples": [ { "input": "The paragraph describes the event of Bristol Myers Squibb reporting their quarterly earnings.", "label": "entailment" }, { "input": "The paragraph describes a corporate merger event involving Bristol Myers Squibb.", "label": "non-entailment" } ] }, { "task": "Impact Classification", "examples": [ { "input": "The paragraph highlights the negative impact of generic competition on Revlimid sales.", "label": "entailment" }, { "input": "The paragraph highlights an increase in Revlimid sales due to increased market share.", "label": "non-entailment" } ] }, { "task": "Tense Identification", "examples": [ { "input": "The paragraph is presented in the past tense, describing events that have already occurred.", "label": "entailment" }, { "input": "The paragraph is presented in the future tense, describing expected events.", "label": "non-entailment" } ] } ] }