Ka whakamātautau te kaikōrero

Five things, depending on the mode you pick: (1) overall polarity - a -100 to +100 score with confidence; (2) 6-dimensional emotions - joy, anger, fear, sadness, surprise, disgust; (3) aspect-based - extracts named features or entities and scores each separately; (4) sarcasm / subtext - flips surface polarity on detected irony; (5) per-sentence - tags every sentence individually. Pick the mode that matches your use case.

He - he arotakenga kōrero 200-wāhi e whakahaere ana i te tauira Qwen 3 30B pūmau i ngā tohu ~200, e āhei ana ki roto i te 6,000 whakawātea, 30,000 rānei i te pūtea i te rā. Ko te tātaritanga āhua-roto i runga i ngā kupu roa e utu ana i ngā tohu ~600. Kāore e hiahiatia te whakamātautau ki ētahi.

E ōrite ana ki te pūtake katoa (~85% whakaaetanga ki ngā kaikōrero tangata i runga i te kupu whakawhāiti mārama), i muri i Google i te tika o te whakawāteatanga kaupapa, i mua i te nuinga o te whakamātautau nā te mea ka hoatu e tātau te tauira i ngā whakahau whakamātautau mō te kitenga o te whakamātautau. Ko ngā kaipāpāho o te kamupene e utu ana $1-4 i ia wa i ngā wae 1,000. Ko tātau painga: te pūwhitinga āhua + te huaputa i runga i te āhua i roto i tētahi whakarongo, wātea, kāore he pūtātau API hei whakarato.

Yes - switch to the "Sarcasm + subtext" mode. The system prompt explicitly instructs the model to flip surface polarity when it detects irony ("Oh great, another delay" scores negative despite positive-word presence). Accuracy on obvious sarcasm is ~80%; subtle passive-aggression is harder for any model. The "Mixed" verdict + low-confidence score is your flag to read the text yourself.

Instead of one verdict for the whole text, you get separate scores for each entity or feature mentioned. Example: "The hotel was beautiful but the breakfast was cold and the wifi was terrible" yields {hotel: +80, breakfast: -50, wifi: -70}. Ideal for product reviews, restaurant feedback, and NPS comments where customers mix praise and complaints in one paragraph.

99 languages. The six emotion dimensions come from Ekman basic-emotion research and translate universally. Accuracy is highest on English, Spanish, French, Portuguese, German, Chinese, Japanese; lower-resource languages work but may produce slightly lower confidence scores.

He - ko taua tētahi o ōna take whakamahi nui rawa. Ka haere ngā tākete ki roto i te āhua rohe "Tuki i te tākete"; ngā tākete e whakawātea ana i raro i te -50 polarity me te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te āhua o te

He - POST ki /v1/chat/ me te whakahau pūnaha ōrite e hanga ana tēnei pātaka (tirohia te pūtake mō te whakahau tika). Ka hoki mai te JSON hanganga. Pai mo ngā tātaitai e whakamātau ana i ngā mano o ngā arotakenga i ia rā. Whakatika te mana, ngā tepe o te marama. Ka whakakōrerotia i te /api/.

Ko te whakapono e whakaatu ana i te mōhiotanga tauira, ehara i te tōmu mōhiotanga. Ko ngā tāuru poto (<20 kupu), te kupu whakamātautau, me ngā ihirangi ihirangi-pāpāhotanga he iti. Ko te tōrunga o te whakapono iti, ehara i te tōraro i te whakapono - ko "kore e mōhio ana au". Ka whakamātau i ngā pūkete i raro i te 50 hei hiahiatia he arotakenga tangata.

Ka haere te UI ki tētahi kupu i tētahi wā. Mō te nui, ka taea e te whakataki JSON+CSV te whakarerekē i te API ki tētahi Python, ki tētahi tuhipānui Node rānei e whakarerekē ana i tō tātau pūkete raraunga. Te whakawhāiti: 20,000 ngā āhuatanga i ia whakarongo, 100 ngā whakarongo/ora i ia IP mō ngā kaiwhakaari ā-whakaaro, 1,000/ora mō ngā kāwanatanga i whakarārangitia. Ka taea ngā tepe o te utu o te kamupene - tātau ki a tātau.

I whiwhi a MonkeyLearn, ā, i whakamutua. Ko Lexalytics me Repustate e utu ana i te $500-5,000 / mo mō te kitenga āhuahira + āhuahira ōrite. He wātea tātau utauta mō te nuinga o ngā rerenga mahi; ngā tauira utu (GPT-5, Claude Sonnet) e wātea ana i ia whakarongo mō ngā take whakamahi tino nui pēnei i ngā whakarerekētanga whakarongo.

Ka tuwhera te kupu i roto i te pūmahara i runga i tātau GPU, kāore i te mau ki te pūrere mō ngā kaiwhakaari ā-pūrere. Ka kitea e ngā kaiwhakaari whakawhiwhia ngā pūrongo i roto i to rātau papatono mō te 7 rā. Kāore anō a tātau kia tiritiri i te kupu ki ngā wāhanga tuatoru, ki te whakamahi rānei mō te whakaakoranga. Mō te whakaritenga pūkete kore, ka whakamahia e tātau tō tātau pūtātau API-key-gated - tātau ki a mātau mō ngā SLAs kamupene.