Rudimentary word-based reply generation

config.defaults.json

@@ -14,6 +14,7 @@
 	"overlap_ratio_enabled": false,
 	"overlap_ratio": 0.7,
 	"generation_mode": "markov",
+	"keyword_from_reply": true,
 	"access_token": "",
 	"db_path": "posts.db"
 }

generators/markov.py

@@ -2,9 +2,9 @@
 
 import sqlite3
 import markovify
-from random import randint
+from random import randint, choice
 
-def make_sentence(cfg):
+def make_sentence(cfg, keywords):
 	class nlt_fixed(markovify.NewlineText):  # modified version of NewlineText that never rejects sentences
 		def test_sentence_input(self, sentence):
 			return True  # all sentences are valid <3
@@ -49,10 +49,25 @@ def make_sentence(cfg):
 	if cfg['limit_length']:
 		sentence_len = randint(cfg['length_lower_limit'], cfg['length_upper_limit'])
 
+	def make_short_sentence_with_keyword(max_chars, min_chars=0, keywords=None, **kwargs):
+		tries = kwargs.get("tries")
+		for _ in range(tries):
+			if keywords:
+				try:
+					keyword = choice(model.word_split(keywords))
+					sentence = model.make_sentence_with_start(keyword, strict=False, **kwargs)
+				except:
+					sentence = model.make_sentence(**kwargs)
+			else:
+				sentence = model.make_sentence(**kwargs)
+			if sentence and min_chars <= len(sentence) <= max_chars:
+				return sentence
+
 	sentence = None
 	tries = 0
 	for tries in range(10):
-		if (sentence := model.make_short_sentence(
+		if (sentence := make_short_sentence_with_keyword(
+			keywords=keywords if cfg['keywords_from_reply'] else None,
 			max_chars=500,
 			tries=10000,
 			max_overlap_ratio=cfg['overlap_ratio'] if cfg['overlap_ratio_enabled'] else 0.7,

pleroma.py

@@ -66,7 +66,7 @@ class Pleroma:
 		id = self._unpack_id(id)
 		return await self.request('GET', f'/api/v1/statuses/{id}/context')
 
-	async def post(self, content, *, in_reply_to_id=None, cw=None, visibility=None):
+	async def post(self, content, *, in_reply_to_id=None, cw=None, visibility=None, keywords=None):
 		if visibility not in {None, 'private', 'public', 'unlisted', 'direct'}:
 			raise ValueError('invalid visibility', visibility)
 

reply.py

@@ -77,9 +77,14 @@ class ReplyBot:
 			if command in ('pin', 'unpin'):
 				await (self.pleroma.pin if command == 'pin' else self.pleroma.unpin)(target_post_id)
 			elif command == 'reply':
-				toot = await utils.make_post(self.cfg)
-				toot = self.cleanup_toot(toot, self.cfg)
 				status = await self.pleroma.get_status(argument)
+				if status['content'] != "":
+					keywords = cleanup_toot(utils.extract_post_content(status['content']), self.cfg)
+				else:
+					keywords = None
+
+				toot = await utils.make_post(self.cfg, keywords)
+				toot = self.cleanup_toot(toot, self.cfg)
 				await self.pleroma.reply(status, toot, cw=self.cfg['cw'])
 		except pleroma.BadRequest as exc:
 			async with anyio.create_task_group() as tg:
@@ -89,9 +94,15 @@ class ReplyBot:
 			await self.pleroma.react(post_id, '✅')
 
 	async def reply(self, notification):
-		toot = await utils.make_post(self.cfg)
+		status = notification['status']
+		if status['content'] != "":
+			keywords = self.cleanup_toot(utils.extract_post_content(status['content']), self.cfg)
+		else:
+			keywords = None
+
+		toot = await utils.make_post(self.cfg, keywords)
 		toot = self.cleanup_toot(toot, self.cfg)
-		await self.pleroma.reply(notification['status'], toot, cw=self.cfg['cw'])
+		await self.pleroma.reply(status, toot, cw=self.cfg['cw'])
 
 	@staticmethod
 	def cleanup_toot(text, cfg):

third_party/utils.py

@@ -63,13 +63,13 @@ def remove_mentions(cfg, sentence):
 
 	return sentence
 
-async def make_post(cfg, *, mode=TextGenerationMode.markov):
+async def make_post(cfg, keywords=None, *, mode=TextGenerationMode.markov):
 	if mode is TextGenerationMode.markov:
 		from generators.markov import make_sentence
 	elif mode is TextGenerationMode.gpt_2:
 		from generators.gpt_2 import make_sentence
 
-	return await anyio.to_process.run_sync(make_sentence, cfg)
+	return await anyio.to_process.run_sync(make_sentence, cfg, keywords)
 
 def extract_post_content(text):
 	soup = BeautifulSoup(text, "html.parser")