我正在尝试理解在使用QuoteAnnotator对新闻语料库进行处理时，我们得到的一些精度较低的结果。我希望了解并调整筛选器以提高精度，但在通过HTTP查询服务器时无法检索到提及和筛选数据。然而，当我使用sample server jupyter notebook时，我确实可以获取这些数据。

CoreNLPClient的结果

我在笔记本中运行了一个测试，如下所示：

# create server
os.environ["CORENLP_HOME"] = "./corenlp"
client = CoreNLPClient(annotators=['tokenize','ssplit', 'pos', 'lemma', 'ner', 'depparse', 'coref', 'quote'], memory='8G', endpoint='http://localhost:9001')
client.start()
time.sleep(10)
text = """Vice President Mike Pence announced Thursday that Israel's Prime Minister and opposition leader will visit the White House next week to discuss "regional issues as well as the prospect of peace." The announcement comes as reports suggest a potential reveal of the Trump administration's Middle East peace plan could be imminent. Pence, who was in Jerusalem for a gathering of world leaders to mark the 75th anniversary of the liberation of Auschwitz, extended the invitation to Prime Minister Benjamin Netanyahu from President Donald Trump. He also announced that Blue and White chairman Benny Gantz will also attend the meeting at the White House on Tuesday. In a tweet Trump appeared to dash speculation an announcement on the peace plan may be imminent. "The United States looks forward to welcoming Prime Minister @Netanyahu & Blue & White Chairman @Gantzbe to the @WhiteHouse next week. Reports about details and timing of our closely-held peace plan are purely speculative," he tweeted. The unveiling of the plan, which is being spearheaded by Trump's senior adviser and son-in-law Jared Kushner, has been delayed amid the months-long period of turmoil in Israeli politics with the country due to hold an unprecedented third national election in less than a year in March."""
document = client.annotate(text)
document.quote[1]

收到了包括mention、mentionSieve和其他属性的结果：

...
mention: "he"
mentionBegin: 171
mentionEnd: 171
mentionType: "pronoun"
mentionSieve: "trigram QPV"
canonicalMention: "Donald Trump"
...

HTTP请求的结果

在同一台笔记本中，查询相同的服务器，我通过HTTP获取json结果，但没有提及信息：

import requests
url = 'http://localhost:9001/?properties={"annotators":"tokenize,ssplit,pos,lemma,ner,depparse,coref,quote","outputFormat":"json"}'
r = requests.post(url, data=text.encode('utf-8'))
results = r.json()
results['quotes'][1]

我得到了以下结果：

{'id': 1,
 'text': '"The United States looks forward to welcoming Prime Minister @Netanyahu & Blue & White Chairman @Gantzbe to the @WhiteHouse next week. Reports about details and timing of our closely-held peace plan are purely speculative,"',
 'beginIndex': 757,
 'endIndex': 980,
 'beginToken': 133,
 'endToken': 170,
 'beginSentence': 5,
 'endSentence': 6,
 'speaker': 'Unknown',
 'canonicalSpeaker': 'Donald Trump'}

问题

我认为我在两种情况下都在做同样的请求。一个线索是，我认为输出格式可能不同- json与serialized。但是为什么服务器会根据指定的输出格式返回不同的结果呢？有什么方法可以在json输出中包含提及数据？我需要直接接入Java代码来实现这一点吗？(可能与#616有关？)